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 usec_t event_timeout_warn_usec = 180 * USEC_PER_SEC / 3;
  79 static sigset_t sigmask_orig;
  80 static UDEV_LIST(event_list);
  81 static UDEV_LIST(worker_list);
  82 static char *udev_cgroup;
  83 static bool udev_exit;
  84
  85 enum event_state {
  86         EVENT_UNDEF,
  87         EVENT_QUEUED,
  88         EVENT_RUNNING,
  89 };
  90
  91 struct event {
  92         struct udev_list_node node;
  93         struct udev *udev;
  94         struct udev_device *dev;
  95         enum event_state state;
  96         int exitcode;
  97         unsigned long long int delaying_seqnum;
  98         unsigned long long int seqnum;
  99         const char *devpath;
 100         size_t devpath_len;
 101         const char *devpath_old;
 102         dev_t devnum;
 103         int ifindex;
 104         bool is_block;
 105 };
 106
 107 static inline struct event *node_to_event(struct udev_list_node *node) {
 108         return container_of(node, struct event, node);
 109 }
 110
 111 static void event_queue_cleanup(struct udev *udev, enum event_state type);
 112
 113 enum worker_state {
 114         WORKER_UNDEF,
 115         WORKER_RUNNING,
 116         WORKER_IDLE,
 117         WORKER_KILLED,
 118 };
 119
 120 struct worker {
 121         struct udev_list_node node;
 122         struct udev *udev;
 123         int refcount;
 124         pid_t pid;
 125         struct udev_monitor *monitor;
 126         enum worker_state state;
 127         struct event *event;
 128         usec_t event_start_usec;
 129         bool event_warned;
 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                 _cleanup_rtnl_unref_ sd_rtnl *rtnl = NULL;
 207                 struct epoll_event ep_signal, ep_monitor;
 208                 sigset_t mask;
 209                 int rc = EXIT_SUCCESS;
 210
 211                 /* take initial device from queue */
 212                 dev = event->dev;
 213                 event->dev = NULL;
 214
 215                 free(worker);
 216                 worker_list_cleanup(udev);
 217                 event_queue_cleanup(udev, EVENT_UNDEF);
 218                 udev_monitor_unref(monitor);
 219                 udev_ctrl_unref(udev_ctrl);
 220                 close(fd_signal);
 221                 close(fd_ep);
 222                 close(worker_watch[READ_END]);
 223
 224                 sigfillset(&mask);
 225                 fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
 226                 if (fd_signal < 0) {
 227                         log_error("error creating signalfd %m");
 228                         rc = 2;
 229                         goto out;
 230                 }
 231
 232                 fd_ep = epoll_create1(EPOLL_CLOEXEC);
 233                 if (fd_ep < 0) {
 234                         log_error("error creating epoll fd: %m");
 235                         rc = 3;
 236                         goto out;
 237                 }
 238
 239                 memzero(&ep_signal, sizeof(struct epoll_event));
 240                 ep_signal.events = EPOLLIN;
 241                 ep_signal.data.fd = fd_signal;
 242
 243                 fd_monitor = udev_monitor_get_fd(worker_monitor);
 244                 memzero(&ep_monitor, sizeof(struct epoll_event));
 245                 ep_monitor.events = EPOLLIN;
 246                 ep_monitor.data.fd = fd_monitor;
 247
 248                 if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
 249                     epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_monitor, &ep_monitor) < 0) {
 250                         log_error("fail to add fds to epoll: %m");
 251                         rc = 4;
 252                         goto out;
 253                 }
 254
 255                 /* request TERM signal if parent exits */
 256                 prctl(PR_SET_PDEATHSIG, SIGTERM);
 257
 258                 /* reset OOM score, we only protect the main daemon */
 259                 write_string_file("/proc/self/oom_score_adj", "0");
 260
 261                 for (;;) {
 262                         struct udev_event *udev_event;
 263                         struct worker_message msg;
 264                         int fd_lock = -1;
 265                         int err = 0;
 266
 267                         log_debug("seq %llu running", udev_device_get_seqnum(dev));
 268                         udev_event = udev_event_new(dev);
 269                         if (udev_event == NULL) {
 270                                 rc = 5;
 271                                 goto out;
 272                         }
 273
 274                         /* needed for SIGCHLD/SIGTERM in spawn() */
 275                         udev_event->fd_signal = fd_signal;
 276
 277                         if (exec_delay > 0)
 278                                 udev_event->exec_delay = exec_delay;
 279
 280                         /*
 281                          * Take a shared lock on the device node; this establishes
 282                          * a concept of device "ownership" to serialize device
 283                          * access. External processes holding an exclusive lock will
 284                          * cause udev to skip the event handling; in the case udev
 285                          * acquired the lock, the external process can block until
 286                          * udev has finished its event handling.
 287                          */
 288                         if (!streq_ptr(udev_device_get_action(dev), "remove") &&
 289                             streq_ptr("block", udev_device_get_subsystem(dev)) &&
 290                             !startswith(udev_device_get_sysname(dev), "dm-") &&
 291                             !startswith(udev_device_get_sysname(dev), "md")) {
 292                                 struct udev_device *d = dev;
 293
 294                                 if (streq_ptr("partition", udev_device_get_devtype(d)))
 295                                         d = udev_device_get_parent(d);
 296
 297                                 if (d) {
 298                                         fd_lock = open(udev_device_get_devnode(d), O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
 299                                         if (fd_lock >= 0 && flock(fd_lock, LOCK_SH|LOCK_NB) < 0) {
 300                                                 log_debug("Unable to flock(%s), skipping event handling: %m", udev_device_get_devnode(d));
 301                                                 err = -EWOULDBLOCK;
 302                                                 fd_lock = safe_close(fd_lock);
 303                                                 goto skip;
 304                                         }
 305                                 }
 306                         }
 307
 308                         /* needed for renaming netifs */
 309                         udev_event->rtnl = rtnl;
 310
 311                         /* apply rules, create node, symlinks */
 312                         udev_event_execute_rules(udev_event, event_timeout_usec, event_timeout_warn_usec, rules, &sigmask_orig);
 313
 314                         udev_event_execute_run(udev_event, event_timeout_usec, event_timeout_warn_usec, &sigmask_orig);
 315
 316                         /* in case rtnl was initialized */
 317                         rtnl = sd_rtnl_ref(udev_event->rtnl);
 318
 319                         /* apply/restore inotify watch */
 320                         if (udev_event->inotify_watch) {
 321                                 udev_watch_begin(udev, dev);
 322                                 udev_device_update_db(dev);
 323                         }
 324
 325                         safe_close(fd_lock);
 326
 327                         /* send processed event back to libudev listeners */
 328                         udev_monitor_send_device(worker_monitor, NULL, dev);
 329
 330 skip:
 331                         /* send udevd the result of the event execution */
 332                         memzero(&msg, sizeof(struct worker_message));
 333                         msg.exitcode = err;
 334                         msg.pid = getpid();
 335                         send(worker_watch[WRITE_END], &msg, sizeof(struct worker_message), 0);
 336
 337                         log_debug("seq %llu processed with %i", udev_device_get_seqnum(dev), err);
 338
 339                         udev_device_unref(dev);
 340                         dev = NULL;
 341
 342                         if (udev_event->sigterm) {
 343                                 udev_event_unref(udev_event);
 344                                 goto out;
 345                         }
 346
 347                         udev_event_unref(udev_event);
 348
 349                         /* wait for more device messages from main udevd, or term signal */
 350                         while (dev == NULL) {
 351                                 struct epoll_event ev[4];
 352                                 int fdcount;
 353                                 int i;
 354
 355                                 fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), -1);
 356                                 if (fdcount < 0) {
 357                                         if (errno == EINTR)
 358                                                 continue;
 359                                         log_error("failed to poll: %m");
 360                                         goto out;
 361                                 }
 362
 363                                 for (i = 0; i < fdcount; i++) {
 364                                         if (ev[i].data.fd == fd_monitor && ev[i].events & EPOLLIN) {
 365                                                 dev = udev_monitor_receive_device(worker_monitor);
 366                                                 break;
 367                                         } else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN) {
 368                                                 struct signalfd_siginfo fdsi;
 369                                                 ssize_t size;
 370
 371                                                 size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
 372                                                 if (size != sizeof(struct signalfd_siginfo))
 373                                                         continue;
 374                                                 switch (fdsi.ssi_signo) {
 375                                                 case SIGTERM:
 376                                                         goto out;
 377                                                 }
 378                                         }
 379                                 }
 380                         }
 381                 }
 382 out:
 383                 udev_device_unref(dev);
 384                 safe_close(fd_signal);
 385                 safe_close(fd_ep);
 386                 close(fd_inotify);
 387                 close(worker_watch[WRITE_END]);
 388                 udev_rules_unref(rules);
 389                 udev_builtin_exit(udev);
 390                 udev_monitor_unref(worker_monitor);
 391                 udev_unref(udev);
 392                 log_close();
 393                 exit(rc);
 394         }
 395         case -1:
 396                 udev_monitor_unref(worker_monitor);
 397                 event->state = EVENT_QUEUED;
 398                 free(worker);
 399                 log_error("fork of child failed: %m");
 400                 break;
 401         default:
 402                 /* close monitor, but keep address around */
 403                 udev_monitor_disconnect(worker_monitor);
 404                 worker->monitor = worker_monitor;
 405                 worker->pid = pid;
 406                 worker->state = WORKER_RUNNING;
 407                 worker->event_start_usec = now(CLOCK_MONOTONIC);
 408                 worker->event_warned = false;
 409                 worker->event = event;
 410                 event->state = EVENT_RUNNING;
 411                 udev_list_node_append(&worker->node, &worker_list);
 412                 children++;
 413                 log_debug("seq %llu forked new worker [%u]", udev_device_get_seqnum(event->dev), pid);
 414                 break;
 415         }
 416 }
 417
 418 static void event_run(struct event *event) {
 419         struct udev_list_node *loop;
 420
 421         udev_list_node_foreach(loop, &worker_list) {
 422                 struct worker *worker = node_to_worker(loop);
 423                 ssize_t count;
 424
 425                 if (worker->state != WORKER_IDLE)
 426                         continue;
 427
 428                 count = udev_monitor_send_device(monitor, worker->monitor, event->dev);
 429                 if (count < 0) {
 430                         log_error("worker [%u] did not accept message %zi (%m), kill it", worker->pid, count);
 431                         kill(worker->pid, SIGKILL);
 432                         worker->state = WORKER_KILLED;
 433                         continue;
 434                 }
 435                 worker_ref(worker);
 436                 worker->event = event;
 437                 worker->state = WORKER_RUNNING;
 438                 worker->event_start_usec = now(CLOCK_MONOTONIC);
 439                 worker->event_warned = false;
 440                 event->state = EVENT_RUNNING;
 441                 return;
 442         }
 443
 444         if (children >= children_max) {
 445                 if (children_max > 1)
 446                         log_debug("maximum number (%i) of children reached", children);
 447                 return;
 448         }
 449
 450         /* start new worker and pass initial device */
 451         worker_new(event);
 452 }
 453
 454 static int event_queue_insert(struct udev_device *dev) {
 455         struct event *event;
 456
 457         event = new0(struct event, 1);
 458         if (event == NULL)
 459                 return -1;
 460
 461         event->udev = udev_device_get_udev(dev);
 462         event->dev = dev;
 463         event->seqnum = udev_device_get_seqnum(dev);
 464         event->devpath = udev_device_get_devpath(dev);
 465         event->devpath_len = strlen(event->devpath);
 466         event->devpath_old = udev_device_get_devpath_old(dev);
 467         event->devnum = udev_device_get_devnum(dev);
 468         event->is_block = streq("block", udev_device_get_subsystem(dev));
 469         event->ifindex = udev_device_get_ifindex(dev);
 470
 471         log_debug("seq %llu queued, '%s' '%s'", udev_device_get_seqnum(dev),
 472              udev_device_get_action(dev), udev_device_get_subsystem(dev));
 473
 474         event->state = EVENT_QUEUED;
 475         udev_list_node_append(&event->node, &event_list);
 476         return 0;
 477 }
 478
 479 static void worker_kill(struct udev *udev) {
 480         struct udev_list_node *loop;
 481
 482         udev_list_node_foreach(loop, &worker_list) {
 483                 struct worker *worker = node_to_worker(loop);
 484
 485                 if (worker->state == WORKER_KILLED)
 486                         continue;
 487
 488                 worker->state = WORKER_KILLED;
 489                 kill(worker->pid, SIGTERM);
 490         }
 491 }
 492
 493 /* lookup event for identical, parent, child device */
 494 static bool is_devpath_busy(struct event *event) {
 495         struct udev_list_node *loop;
 496         size_t common;
 497
 498         /* check if queue contains events we depend on */
 499         udev_list_node_foreach(loop, &event_list) {
 500                 struct event *loop_event = node_to_event(loop);
 501
 502                 /* we already found a later event, earlier can not block us, no need to check again */
 503                 if (loop_event->seqnum < event->delaying_seqnum)
 504                         continue;
 505
 506                 /* event we checked earlier still exists, no need to check again */
 507                 if (loop_event->seqnum == event->delaying_seqnum)
 508                         return true;
 509
 510                 /* found ourself, no later event can block us */
 511                 if (loop_event->seqnum >= event->seqnum)
 512                         break;
 513
 514                 /* check major/minor */
 515                 if (major(event->devnum) != 0 && event->devnum == loop_event->devnum && event->is_block == loop_event->is_block)
 516                         return true;
 517
 518                 /* check network device ifindex */
 519                 if (event->ifindex != 0 && event->ifindex == loop_event->ifindex)
 520                         return true;
 521
 522                 /* check our old name */
 523                 if (event->devpath_old != NULL && streq(loop_event->devpath, event->devpath_old)) {
 524                         event->delaying_seqnum = loop_event->seqnum;
 525                         return true;
 526                 }
 527
 528                 /* compare devpath */
 529                 common = MIN(loop_event->devpath_len, event->devpath_len);
 530
 531                 /* one devpath is contained in the other? */
 532                 if (memcmp(loop_event->devpath, event->devpath, common) != 0)
 533                         continue;
 534
 535                 /* identical device event found */
 536                 if (loop_event->devpath_len == event->devpath_len) {
 537                         /* devices names might have changed/swapped in the meantime */
 538                         if (major(event->devnum) != 0 && (event->devnum != loop_event->devnum || event->is_block != loop_event->is_block))
 539                                 continue;
 540                         if (event->ifindex != 0 && event->ifindex != loop_event->ifindex)
 541                                 continue;
 542                         event->delaying_seqnum = loop_event->seqnum;
 543                         return true;
 544                 }
 545
 546                 /* parent device event found */
 547                 if (event->devpath[common] == '/') {
 548                         event->delaying_seqnum = loop_event->seqnum;
 549                         return true;
 550                 }
 551
 552                 /* child device event found */
 553                 if (loop_event->devpath[common] == '/') {
 554                         event->delaying_seqnum = loop_event->seqnum;
 555                         return true;
 556                 }
 557
 558                 /* no matching device */
 559                 continue;
 560         }
 561
 562         return false;
 563 }
 564
 565 static void event_queue_start(struct udev *udev) {
 566         struct udev_list_node *loop;
 567
 568         udev_list_node_foreach(loop, &event_list) {
 569                 struct event *event = node_to_event(loop);
 570
 571                 if (event->state != EVENT_QUEUED)
 572                         continue;
 573
 574                 /* do not start event if parent or child event is still running */
 575                 if (is_devpath_busy(event))
 576                         continue;
 577
 578                 event_run(event);
 579         }
 580 }
 581
 582 static void event_queue_cleanup(struct udev *udev, enum event_state match_type) {
 583         struct udev_list_node *loop, *tmp;
 584
 585         udev_list_node_foreach_safe(loop, tmp, &event_list) {
 586                 struct event *event = node_to_event(loop);
 587
 588                 if (match_type != EVENT_UNDEF && match_type != event->state)
 589                         continue;
 590
 591                 event_queue_delete(event);
 592         }
 593 }
 594
 595 static void worker_returned(int fd_worker) {
 596         for (;;) {
 597                 struct worker_message msg;
 598                 ssize_t size;
 599                 struct udev_list_node *loop;
 600
 601                 size = recv(fd_worker, &msg, sizeof(struct worker_message), MSG_DONTWAIT);
 602                 if (size != sizeof(struct worker_message))
 603                         break;
 604
 605                 /* lookup worker who sent the signal */
 606                 udev_list_node_foreach(loop, &worker_list) {
 607                         struct worker *worker = node_to_worker(loop);
 608
 609                         if (worker->pid != msg.pid)
 610                                 continue;
 611
 612                         /* worker returned */
 613                         if (worker->event) {
 614                                 worker->event->exitcode = msg.exitcode;
 615                                 event_queue_delete(worker->event);
 616                                 worker->event = NULL;
 617                         }
 618                         if (worker->state != WORKER_KILLED)
 619                                 worker->state = WORKER_IDLE;
 620                         worker_unref(worker);
 621                         break;
 622                 }
 623         }
 624 }
 625
 626 /* receive the udevd message from userspace */
 627 static struct udev_ctrl_connection *handle_ctrl_msg(struct udev_ctrl *uctrl) {
 628         struct udev *udev = udev_ctrl_get_udev(uctrl);
 629         struct udev_ctrl_connection *ctrl_conn;
 630         struct udev_ctrl_msg *ctrl_msg = NULL;
 631         const char *str;
 632         int i;
 633
 634         ctrl_conn = udev_ctrl_get_connection(uctrl);
 635         if (ctrl_conn == NULL)
 636                 goto out;
 637
 638         ctrl_msg = udev_ctrl_receive_msg(ctrl_conn);
 639         if (ctrl_msg == NULL)
 640                 goto out;
 641
 642         i = udev_ctrl_get_set_log_level(ctrl_msg);
 643         if (i >= 0) {
 644                 log_debug("udevd message (SET_LOG_LEVEL) received, log_priority=%i", i);
 645                 log_set_max_level(i);
 646                 udev_set_log_priority(udev, i);
 647                 worker_kill(udev);
 648         }
 649
 650         if (udev_ctrl_get_stop_exec_queue(ctrl_msg) > 0) {
 651                 log_debug("udevd message (STOP_EXEC_QUEUE) received");
 652                 stop_exec_queue = true;
 653         }
 654
 655         if (udev_ctrl_get_start_exec_queue(ctrl_msg) > 0) {
 656                 log_debug("udevd message (START_EXEC_QUEUE) received");
 657                 stop_exec_queue = false;
 658         }
 659
 660         if (udev_ctrl_get_reload(ctrl_msg) > 0) {
 661                 log_debug("udevd message (RELOAD) received");
 662                 reload = true;
 663         }
 664
 665         str = udev_ctrl_get_set_env(ctrl_msg);
 666         if (str != NULL) {
 667                 char *key;
 668
 669                 key = strdup(str);
 670                 if (key != NULL) {
 671                         char *val;
 672
 673                         val = strchr(key, '=');
 674                         if (val != NULL) {
 675                                 val[0] = '\0';
 676                                 val = &val[1];
 677                                 if (val[0] == '\0') {
 678                                         log_debug("udevd message (ENV) received, unset '%s'", key);
 679                                         udev_add_property(udev, key, NULL);
 680                                 } else {
 681                                         log_debug("udevd message (ENV) received, set '%s=%s'", key, val);
 682                                         udev_add_property(udev, key, val);
 683                                 }
 684                         } else {
 685                                 log_error("wrong key format '%s'", key);
 686                         }
 687                         free(key);
 688                 }
 689                 worker_kill(udev);
 690         }
 691
 692         i = udev_ctrl_get_set_children_max(ctrl_msg);
 693         if (i >= 0) {
 694                 log_debug("udevd message (SET_MAX_CHILDREN) received, children_max=%i", i);
 695                 children_max = i;
 696         }
 697
 698         if (udev_ctrl_get_ping(ctrl_msg) > 0)
 699                 log_debug("udevd message (SYNC) received");
 700
 701         if (udev_ctrl_get_exit(ctrl_msg) > 0) {
 702                 log_debug("udevd message (EXIT) received");
 703                 udev_exit = true;
 704                 /* keep reference to block the client until we exit */
 705                 udev_ctrl_connection_ref(ctrl_conn);
 706         }
 707 out:
 708         udev_ctrl_msg_unref(ctrl_msg);
 709         return udev_ctrl_connection_unref(ctrl_conn);
 710 }
 711
 712 static int synthesize_change(struct udev_device *dev) {
 713         char filename[UTIL_PATH_SIZE];
 714         int r;
 715
 716         if (streq_ptr("block", udev_device_get_subsystem(dev)) &&
 717             streq_ptr("disk", udev_device_get_devtype(dev)) &&
 718             !startswith(udev_device_get_sysname(dev), "dm-")) {
 719                 bool part_table_read = false;
 720                 bool has_partitions = false;
 721                 int fd;
 722                 struct udev *udev = udev_device_get_udev(dev);
 723                 _cleanup_udev_enumerate_unref_ struct udev_enumerate *e = NULL;
 724                 struct udev_list_entry *item;
 725
 726                 /*
 727                  * Try to re-read the partition table. This only succeeds if
 728                  * none of the devices is busy. The kernel returns 0 if no
 729                  * partition table is found, and we will not get an event for
 730                  * the disk.
 731                  */
 732                 fd = open(udev_device_get_devnode(dev), O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
 733                 if (fd >= 0) {
 734                         r = flock(fd, LOCK_EX|LOCK_NB);
 735                         if (r >= 0)
 736                                 r = ioctl(fd, BLKRRPART, 0);
 737
 738                         close(fd);
 739                         if (r >= 0)
 740                                 part_table_read = true;
 741                 }
 742
 743                 /* search for partitions */
 744                 e = udev_enumerate_new(udev);
 745                 if (!e)
 746                         return -ENOMEM;
 747
 748                 r = udev_enumerate_add_match_parent(e, dev);
 749                 if (r < 0)
 750                         return r;
 751
 752                 r = udev_enumerate_add_match_subsystem(e, "block");
 753                 if (r < 0)
 754                         return r;
 755
 756                 r = udev_enumerate_scan_devices(e);
 757                 if (r < 0)
 758                         return r;
 759
 760                 udev_list_entry_foreach(item, udev_enumerate_get_list_entry(e)) {
 761                         _cleanup_udev_device_unref_ struct udev_device *d = NULL;
 762
 763                         d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
 764                         if (!d)
 765                                 continue;
 766
 767                         if (!streq_ptr("partition", udev_device_get_devtype(d)))
 768                                 continue;
 769
 770                         has_partitions = true;
 771                         break;
 772                 }
 773
 774                 /*
 775                  * We have partitions and re-read the table, the kernel already sent
 776                  * out a "change" event for the disk, and "remove/add" for all
 777                  * partitions.
 778                  */
 779                 if (part_table_read && has_partitions)
 780                         return 0;
 781
 782                 /*
 783                  * We have partitions but re-reading the partition table did not
 784                  * work, synthesize "change" for the disk and all partitions.
 785                  */
 786                 log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev));
 787                 strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
 788                 write_string_file(filename, "change");
 789
 790                 udev_list_entry_foreach(item, udev_enumerate_get_list_entry(e)) {
 791                         _cleanup_udev_device_unref_ struct udev_device *d = NULL;
 792
 793                         d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
 794                         if (!d)
 795                                 continue;
 796
 797                         if (!streq_ptr("partition", udev_device_get_devtype(d)))
 798                                 continue;
 799
 800                         log_debug("device %s closed, synthesising partition '%s' 'change'",
 801                                   udev_device_get_devnode(dev), udev_device_get_devnode(d));
 802                         strscpyl(filename, sizeof(filename), udev_device_get_syspath(d), "/uevent", NULL);
 803                         write_string_file(filename, "change");
 804                 }
 805
 806                 return 0;
 807         }
 808
 809         log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev));
 810         strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
 811         write_string_file(filename, "change");
 812
 813         return 0;
 814 }
 815
 816 static int handle_inotify(struct udev *udev) {
 817         int nbytes, pos;
 818         char *buf;
 819         struct inotify_event *ev;
 820         int r;
 821
 822         r = ioctl(fd_inotify, FIONREAD, &nbytes);
 823         if (r < 0 || nbytes <= 0)
 824                 return -errno;
 825
 826         buf = malloc(nbytes);
 827         if (!buf) {
 828                 log_error("error getting buffer for inotify");
 829                 return -ENOMEM;
 830         }
 831
 832         nbytes = read(fd_inotify, buf, nbytes);
 833
 834         for (pos = 0; pos < nbytes; pos += sizeof(struct inotify_event) + ev->len) {
 835                 struct udev_device *dev;
 836
 837                 ev = (struct inotify_event *)(buf + pos);
 838                 dev = udev_watch_lookup(udev, ev->wd);
 839                 if (!dev)
 840                         continue;
 841
 842                 log_debug("inotify event: %x for %s", ev->mask, udev_device_get_devnode(dev));
 843                 if (ev->mask & IN_CLOSE_WRITE)
 844                         synthesize_change(dev);
 845                 else if (ev->mask & IN_IGNORED)
 846                         udev_watch_end(udev, dev);
 847
 848                 udev_device_unref(dev);
 849         }
 850
 851         free(buf);
 852         return 0;
 853 }
 854
 855 static void handle_signal(struct udev *udev, int signo) {
 856         switch (signo) {
 857         case SIGINT:
 858         case SIGTERM:
 859                 udev_exit = true;
 860                 break;
 861         case SIGCHLD:
 862                 for (;;) {
 863                         pid_t pid;
 864                         int status;
 865                         struct udev_list_node *loop, *tmp;
 866
 867                         pid = waitpid(-1, &status, WNOHANG);
 868                         if (pid <= 0)
 869                                 break;
 870
 871                         udev_list_node_foreach_safe(loop, tmp, &worker_list) {
 872                                 struct worker *worker = node_to_worker(loop);
 873
 874                                 if (worker->pid != pid)
 875                                         continue;
 876                                 log_debug("worker [%u] exit", pid);
 877
 878                                 if (WIFEXITED(status)) {
 879                                         if (WEXITSTATUS(status) != 0)
 880                                                 log_error("worker [%u] exit with return code %i",
 881                                                           pid, WEXITSTATUS(status));
 882                                 } else if (WIFSIGNALED(status)) {
 883                                         log_error("worker [%u] terminated by signal %i (%s)",
 884                                                   pid, WTERMSIG(status), strsignal(WTERMSIG(status)));
 885                                 } else if (WIFSTOPPED(status)) {
 886                                         log_error("worker [%u] stopped", pid);
 887                                 } else if (WIFCONTINUED(status)) {
 888                                         log_error("worker [%u] continued", pid);
 889                                 } else {
 890                                         log_error("worker [%u] exit with status 0x%04x", pid, status);
 891                                 }
 892
 893                                 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 894                                         if (worker->event) {
 895                                                 log_error("worker [%u] failed while handling '%s'",
 896                                                           pid, worker->event->devpath);
 897                                                 worker->event->exitcode = -32;
 898                                                 event_queue_delete(worker->event);
 899
 900                                                 /* drop reference taken for state 'running' */
 901                                                 worker_unref(worker);
 902                                         }
 903                                 }
 904                                 worker_unref(worker);
 905                                 break;
 906                         }
 907                 }
 908                 break;
 909         case SIGHUP:
 910                 reload = true;
 911                 break;
 912         }
 913 }
 914
 915 static int systemd_fds(struct udev *udev, int *rctrl, int *rnetlink) {
 916         int ctrl = -1, netlink = -1;
 917         int fd, n;
 918
 919         n = sd_listen_fds(true);
 920         if (n <= 0)
 921                 return -1;
 922
 923         for (fd = SD_LISTEN_FDS_START; fd < n + SD_LISTEN_FDS_START; fd++) {
 924                 if (sd_is_socket(fd, AF_LOCAL, SOCK_SEQPACKET, -1)) {
 925                         if (ctrl >= 0)
 926                                 return -1;
 927                         ctrl = fd;
 928                         continue;
 929                 }
 930
 931                 if (sd_is_socket(fd, AF_NETLINK, SOCK_RAW, -1)) {
 932                         if (netlink >= 0)
 933                                 return -1;
 934                         netlink = fd;
 935                         continue;
 936                 }
 937
 938                 return -1;
 939         }
 940
 941         if (ctrl < 0 || netlink < 0)
 942                 return -1;
 943
 944         log_debug("ctrl=%i netlink=%i", ctrl, netlink);
 945         *rctrl = ctrl;
 946         *rnetlink = netlink;
 947         return 0;
 948 }
 949
 950 /*
 951  * read the kernel commandline, in case we need to get into debug mode
 952  *   udev.log-priority=<level>              syslog priority
 953  *   udev.children-max=<number of workers>  events are fully serialized if set to 1
 954  *   udev.exec-delay=<number of seconds>    delay execution of every executed program
 955  */
 956 static void kernel_cmdline_options(struct udev *udev) {
 957         _cleanup_free_ char *line = NULL;
 958         const char *word, *state;
 959         size_t l;
 960         int r;
 961
 962         r = proc_cmdline(&line);
 963         if (r < 0)
 964                 log_warning("Failed to read /proc/cmdline, ignoring: %s", strerror(-r));
 965         if (r <= 0)
 966                 return;
 967
 968         FOREACH_WORD_QUOTED(word, l, line, state) {
 969                 char *s, *opt;
 970
 971                 s = strndup(word, l);
 972                 if (!s)
 973                         break;
 974
 975                 /* accept the same options for the initrd, prefixed with "rd." */
 976                 if (in_initrd() && startswith(s, "rd."))
 977                         opt = s + 3;
 978                 else
 979                         opt = s;
 980
 981                 if (startswith(opt, "udev.log-priority=")) {
 982                         int prio;
 983
 984                         prio = util_log_priority(opt + 18);
 985                         log_set_max_level(prio);
 986                         udev_set_log_priority(udev, prio);
 987                 } else if (startswith(opt, "udev.children-max=")) {
 988                         children_max = strtoul(opt + 18, NULL, 0);
 989                 } else if (startswith(opt, "udev.exec-delay=")) {
 990                         exec_delay = strtoul(opt + 16, NULL, 0);
 991                 } else if (startswith(opt, "udev.event-timeout=")) {
 992                         event_timeout_usec = strtoul(opt + 16, NULL, 0) * USEC_PER_SEC;
 993                         event_timeout_warn_usec = (event_timeout_usec / 3) ? : 1;
 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                         event_timeout_warn_usec = (event_timeout_usec / 3) ? : 1;
1057                         break;
1058                 case 'D':
1059                         debug = true;
1060                         log_set_max_level(LOG_DEBUG);
1061                         udev_set_log_priority(udev, LOG_DEBUG);
1062                         break;
1063                 case 'N':
1064                         if (streq(optarg, "early")) {
1065                                 resolve_names = 1;
1066                         } else if (streq(optarg, "late")) {
1067                                 resolve_names = 0;
1068                         } else if (streq(optarg, "never")) {
1069                                 resolve_names = -1;
1070                         } else {
1071                                 fprintf(stderr, "resolve-names must be early, late or never\n");
1072                                 log_error("resolve-names must be early, late or never");
1073                                 goto exit;
1074                         }
1075                         break;
1076                 case 'h':
1077                         printf("Usage: udevd OPTIONS\n"
1078                                "  --daemon\n"
1079                                "  --debug\n"
1080                                "  --children-max=<maximum number of workers>\n"
1081                                "  --exec-delay=<seconds to wait before executing RUN=>\n"
1082                                "  --event-timeout=<seconds to wait before terminating an event>\n"
1083                                "  --resolve-names=early|late|never\n"
1084                                "  --version\n"
1085                                "  --help\n"
1086                                "\n");
1087                         goto exit;
1088                 case 'V':
1089                         printf("%s\n", VERSION);
1090                         goto exit;
1091                 default:
1092                         goto exit;
1093                 }
1094         }
1095
1096         kernel_cmdline_options(udev);
1097
1098         if (getuid() != 0) {
1099                 fprintf(stderr, "root privileges required\n");
1100                 log_error("root privileges required");
1101                 goto exit;
1102         }
1103
1104         /* set umask before creating any file/directory */
1105         chdir("/");
1106         umask(022);
1107
1108         mkdir("/run/udev", 0755);
1109
1110         dev_setup(NULL);
1111
1112         /* before opening new files, make sure std{in,out,err} fds are in a sane state */
1113         if (daemonize) {
1114                 int fd;
1115
1116                 fd = open("/dev/null", O_RDWR);
1117                 if (fd >= 0) {
1118                         if (write(STDOUT_FILENO, 0, 0) < 0)
1119                                 dup2(fd, STDOUT_FILENO);
1120                         if (write(STDERR_FILENO, 0, 0) < 0)
1121                                 dup2(fd, STDERR_FILENO);
1122                         if (fd > STDERR_FILENO)
1123                                 close(fd);
1124                 } else {
1125                         fprintf(stderr, "cannot open /dev/null\n");
1126                         log_error("cannot open /dev/null");
1127                 }
1128         }
1129
1130         if (systemd_fds(udev, &fd_ctrl, &fd_netlink) >= 0) {
1131                 /* get control and netlink socket from systemd */
1132                 udev_ctrl = udev_ctrl_new_from_fd(udev, fd_ctrl);
1133                 if (udev_ctrl == NULL) {
1134                         log_error("error taking over udev control socket");
1135                         rc = 1;
1136                         goto exit;
1137                 }
1138
1139                 monitor = udev_monitor_new_from_netlink_fd(udev, "kernel", fd_netlink);
1140                 if (monitor == NULL) {
1141                         log_error("error taking over netlink socket");
1142                         rc = 3;
1143                         goto exit;
1144                 }
1145
1146                 /* get our own cgroup, we regularly kill everything udev has left behind */
1147                 if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &udev_cgroup) < 0)
1148                         udev_cgroup = NULL;
1149         } else {
1150                 /* open control and netlink socket */
1151                 udev_ctrl = udev_ctrl_new(udev);
1152                 if (udev_ctrl == NULL) {
1153                         fprintf(stderr, "error initializing udev control socket");
1154                         log_error("error initializing udev control socket");
1155                         rc = 1;
1156                         goto exit;
1157                 }
1158                 fd_ctrl = udev_ctrl_get_fd(udev_ctrl);
1159
1160                 monitor = udev_monitor_new_from_netlink(udev, "kernel");
1161                 if (monitor == NULL) {
1162                         fprintf(stderr, "error initializing netlink socket\n");
1163                         log_error("error initializing netlink socket");
1164                         rc = 3;
1165                         goto exit;
1166                 }
1167                 fd_netlink = udev_monitor_get_fd(monitor);
1168         }
1169
1170         if (udev_monitor_enable_receiving(monitor) < 0) {
1171                 fprintf(stderr, "error binding netlink socket\n");
1172                 log_error("error binding netlink socket");
1173                 rc = 3;
1174                 goto exit;
1175         }
1176
1177         if (udev_ctrl_enable_receiving(udev_ctrl) < 0) {
1178                 fprintf(stderr, "error binding udev control socket\n");
1179                 log_error("error binding udev control socket");
1180                 rc = 1;
1181                 goto exit;
1182         }
1183
1184         udev_monitor_set_receive_buffer_size(monitor, 128 * 1024 * 1024);
1185
1186         if (daemonize) {
1187                 pid_t pid;
1188
1189                 pid = fork();
1190                 switch (pid) {
1191                 case 0:
1192                         break;
1193                 case -1:
1194                         log_error("fork of daemon failed: %m");
1195                         rc = 4;
1196                         goto exit;
1197                 default:
1198                         rc = EXIT_SUCCESS;
1199                         goto exit_daemonize;
1200                 }
1201
1202                 setsid();
1203
1204                 write_string_file("/proc/self/oom_score_adj", "-1000");
1205         } else {
1206                 sd_notify(1, "READY=1");
1207         }
1208
1209         log_info("starting version " VERSION "\n");
1210
1211         if (!debug) {
1212                 int fd;
1213
1214                 fd = open("/dev/null", O_RDWR);
1215                 if (fd >= 0) {
1216                         dup2(fd, STDIN_FILENO);
1217                         dup2(fd, STDOUT_FILENO);
1218                         dup2(fd, STDERR_FILENO);
1219                         close(fd);
1220                 }
1221         }
1222
1223         fd_inotify = udev_watch_init(udev);
1224         if (fd_inotify < 0) {
1225                 fprintf(stderr, "error initializing inotify\n");
1226                 log_error("error initializing inotify");
1227                 rc = 4;
1228                 goto exit;
1229         }
1230         udev_watch_restore(udev);
1231
1232         /* block and listen to all signals on signalfd */
1233         sigfillset(&mask);
1234         sigprocmask(SIG_SETMASK, &mask, &sigmask_orig);
1235         fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
1236         if (fd_signal < 0) {
1237                 fprintf(stderr, "error creating signalfd\n");
1238                 log_error("error creating signalfd");
1239                 rc = 5;
1240                 goto exit;
1241         }
1242
1243         /* unnamed socket from workers to the main daemon */
1244         if (socketpair(AF_LOCAL, SOCK_DGRAM|SOCK_CLOEXEC, 0, worker_watch) < 0) {
1245                 fprintf(stderr, "error creating socketpair\n");
1246                 log_error("error creating socketpair");
1247                 rc = 6;
1248                 goto exit;
1249         }
1250         fd_worker = worker_watch[READ_END];
1251
1252         udev_builtin_init(udev);
1253
1254         rules = udev_rules_new(udev, resolve_names);
1255         if (rules == NULL) {
1256                 log_error("error reading rules");
1257                 goto exit;
1258         }
1259
1260         memzero(&ep_ctrl, sizeof(struct epoll_event));
1261         ep_ctrl.events = EPOLLIN;
1262         ep_ctrl.data.fd = fd_ctrl;
1263
1264         memzero(&ep_inotify, sizeof(struct epoll_event));
1265         ep_inotify.events = EPOLLIN;
1266         ep_inotify.data.fd = fd_inotify;
1267
1268         memzero(&ep_signal, sizeof(struct epoll_event));
1269         ep_signal.events = EPOLLIN;
1270         ep_signal.data.fd = fd_signal;
1271
1272         memzero(&ep_netlink, sizeof(struct epoll_event));
1273         ep_netlink.events = EPOLLIN;
1274         ep_netlink.data.fd = fd_netlink;
1275
1276         memzero(&ep_worker, sizeof(struct epoll_event));
1277         ep_worker.events = EPOLLIN;
1278         ep_worker.data.fd = fd_worker;
1279
1280         fd_ep = epoll_create1(EPOLL_CLOEXEC);
1281         if (fd_ep < 0) {
1282                 log_error("error creating epoll fd: %m");
1283                 goto exit;
1284         }
1285         if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_ctrl, &ep_ctrl) < 0 ||
1286             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_inotify, &ep_inotify) < 0 ||
1287             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
1288             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_netlink, &ep_netlink) < 0 ||
1289             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_worker, &ep_worker) < 0) {
1290                 log_error("fail to add fds to epoll: %m");
1291                 goto exit;
1292         }
1293
1294         if (children_max <= 0) {
1295                 cpu_set_t cpu_set;
1296
1297                 children_max = 8;
1298
1299                 if (sched_getaffinity(0, sizeof (cpu_set), &cpu_set) == 0) {
1300                         children_max +=  CPU_COUNT(&cpu_set) * 2;
1301                 }
1302         }
1303         log_debug("set children_max to %u", children_max);
1304
1305         rc = udev_rules_apply_static_dev_perms(rules);
1306         if (rc < 0)
1307                 log_error("failed to apply permissions on static device nodes - %s", strerror(-rc));
1308
1309         udev_list_node_init(&event_list);
1310         udev_list_node_init(&worker_list);
1311
1312         for (;;) {
1313                 static usec_t last_usec;
1314                 struct epoll_event ev[8];
1315                 int fdcount;
1316                 int timeout;
1317                 bool is_worker, is_signal, is_inotify, is_netlink, is_ctrl;
1318                 int i;
1319
1320                 if (udev_exit) {
1321                         /* close sources of new events and discard buffered events */
1322                         if (fd_ctrl >= 0) {
1323                                 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_ctrl, NULL);
1324                                 fd_ctrl = -1;
1325                         }
1326                         if (monitor != NULL) {
1327                                 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_netlink, NULL);
1328                                 udev_monitor_unref(monitor);
1329                                 monitor = NULL;
1330                         }
1331                         if (fd_inotify >= 0) {
1332                                 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_inotify, NULL);
1333                                 close(fd_inotify);
1334                                 fd_inotify = -1;
1335                         }
1336
1337                         /* discard queued events and kill workers */
1338                         event_queue_cleanup(udev, EVENT_QUEUED);
1339                         worker_kill(udev);
1340
1341                         /* exit after all has cleaned up */
1342                         if (udev_list_node_is_empty(&event_list) && children == 0)
1343                                 break;
1344
1345                         /* timeout at exit for workers to finish */
1346                         timeout = 30 * MSEC_PER_SEC;
1347                 } else if (udev_list_node_is_empty(&event_list) && children == 0) {
1348                         /* we are idle */
1349                         timeout = -1;
1350
1351                         /* cleanup possible left-over processes in our cgroup */
1352                         if (udev_cgroup)
1353                                 cg_kill(SYSTEMD_CGROUP_CONTROLLER, udev_cgroup, SIGKILL, false, true, NULL);
1354                 } else {
1355                         /* kill idle or hanging workers */
1356                         timeout = 3 * MSEC_PER_SEC;
1357                 }
1358
1359                 /* tell settle that we are busy or idle */
1360                 if (!udev_list_node_is_empty(&event_list)) {
1361                         int fd;
1362
1363                         fd = open("/run/udev/queue", O_WRONLY|O_CREAT|O_CLOEXEC|O_TRUNC|O_NOFOLLOW, 0444);
1364                         if (fd >= 0)
1365                                 close(fd);
1366                 } else {
1367                         unlink("/run/udev/queue");
1368                 }
1369
1370                 fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), timeout);
1371                 if (fdcount < 0)
1372                         continue;
1373
1374                 if (fdcount == 0) {
1375                         struct udev_list_node *loop;
1376
1377                         /* timeout */
1378                         if (udev_exit) {
1379                                 log_error("timeout, giving up waiting for workers to finish");
1380                                 break;
1381                         }
1382
1383                         /* kill idle workers */
1384                         if (udev_list_node_is_empty(&event_list)) {
1385                                 log_debug("cleanup idle workers");
1386                                 worker_kill(udev);
1387                         }
1388
1389                         /* check for hanging events */
1390                         udev_list_node_foreach(loop, &worker_list) {
1391                                 struct worker *worker = node_to_worker(loop);
1392                                 usec_t ts;
1393
1394                                 if (worker->state != WORKER_RUNNING)
1395                                         continue;
1396
1397                                 ts = now(CLOCK_MONOTONIC);
1398
1399                                 if ((ts - worker->event_start_usec) > event_timeout_warn_usec) {
1400                                         if ((ts - worker->event_start_usec) > event_timeout_usec) {
1401                                                 log_error("worker [%u] %s timeout; kill it", worker->pid, worker->event->devpath);
1402                                                 kill(worker->pid, SIGKILL);
1403                                                 worker->state = WORKER_KILLED;
1404
1405                                                 /* drop reference taken for state 'running' */
1406                                                 worker_unref(worker);
1407                                                 log_error("seq %llu '%s' killed", udev_device_get_seqnum(worker->event->dev), worker->event->devpath);
1408                                                 worker->event->exitcode = -64;
1409                                                 event_queue_delete(worker->event);
1410                                                 worker->event = NULL;
1411                                         } else if (!worker->event_warned) {
1412                                                 log_warning("worker [%u] %s is taking a long time", worker->pid, worker->event->devpath);
1413                                                 worker->event_warned = true;
1414                                         }
1415                                 }
1416                         }
1417
1418                 }
1419
1420                 is_worker = is_signal = is_inotify = is_netlink = is_ctrl = false;
1421                 for (i = 0; i < fdcount; i++) {
1422                         if (ev[i].data.fd == fd_worker && ev[i].events & EPOLLIN)
1423                                 is_worker = true;
1424                         else if (ev[i].data.fd == fd_netlink && ev[i].events & EPOLLIN)
1425                                 is_netlink = true;
1426                         else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN)
1427                                 is_signal = true;
1428                         else if (ev[i].data.fd == fd_inotify && ev[i].events & EPOLLIN)
1429                                 is_inotify = true;
1430                         else if (ev[i].data.fd == fd_ctrl && ev[i].events & EPOLLIN)
1431                                 is_ctrl = true;
1432                 }
1433
1434                 /* check for changed config, every 3 seconds at most */
1435                 if ((now(CLOCK_MONOTONIC) - last_usec) > 3 * USEC_PER_SEC) {
1436                         if (udev_rules_check_timestamp(rules))
1437                                 reload = true;
1438                         if (udev_builtin_validate(udev))
1439                                 reload = true;
1440
1441                         last_usec = now(CLOCK_MONOTONIC);
1442                 }
1443
1444                 /* reload requested, HUP signal received, rules changed, builtin changed */
1445                 if (reload) {
1446                         worker_kill(udev);
1447                         rules = udev_rules_unref(rules);
1448                         udev_builtin_exit(udev);
1449                         reload = false;
1450                 }
1451
1452                 /* event has finished */
1453                 if (is_worker)
1454                         worker_returned(fd_worker);
1455
1456                 if (is_netlink) {
1457                         struct udev_device *dev;
1458
1459                         dev = udev_monitor_receive_device(monitor);
1460                         if (dev != NULL) {
1461                                 udev_device_set_usec_initialized(dev, now(CLOCK_MONOTONIC));
1462                                 if (event_queue_insert(dev) < 0)
1463                                         udev_device_unref(dev);
1464                         }
1465                 }
1466
1467                 /* start new events */
1468                 if (!udev_list_node_is_empty(&event_list) && !udev_exit && !stop_exec_queue) {
1469                         udev_builtin_init(udev);
1470                         if (rules == NULL)
1471                                 rules = udev_rules_new(udev, resolve_names);
1472                         if (rules != NULL)
1473                                 event_queue_start(udev);
1474                 }
1475
1476                 if (is_signal) {
1477                         struct signalfd_siginfo fdsi;
1478                         ssize_t size;
1479
1480                         size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
1481                         if (size == sizeof(struct signalfd_siginfo))
1482                                 handle_signal(udev, fdsi.ssi_signo);
1483                 }
1484
1485                 /* we are shutting down, the events below are not handled anymore */
1486                 if (udev_exit)
1487                         continue;
1488
1489                 /* device node watch */
1490                 if (is_inotify)
1491                         handle_inotify(udev);
1492
1493                 /*
1494                  * This needs to be after the inotify handling, to make sure,
1495                  * that the ping is send back after the possibly generated
1496                  * "change" events by the inotify device node watch.
1497                  *
1498                  * A single time we may receive a client connection which we need to
1499                  * keep open to block the client. It will be closed right before we
1500                  * exit.
1501                  */
1502                 if (is_ctrl)
1503                         ctrl_conn = handle_ctrl_msg(udev_ctrl);
1504         }
1505
1506         rc = EXIT_SUCCESS;
1507 exit:
1508         udev_ctrl_cleanup(udev_ctrl);
1509         unlink("/run/udev/queue");
1510 exit_daemonize:
1511         if (fd_ep >= 0)
1512                 close(fd_ep);
1513         worker_list_cleanup(udev);
1514         event_queue_cleanup(udev, EVENT_UNDEF);
1515         udev_rules_unref(rules);
1516         udev_builtin_exit(udev);
1517         if (fd_signal >= 0)
1518                 close(fd_signal);
1519         if (worker_watch[READ_END] >= 0)
1520                 close(worker_watch[READ_END]);
1521         if (worker_watch[WRITE_END] >= 0)
1522                 close(worker_watch[WRITE_END]);
1523         udev_monitor_unref(monitor);
1524         udev_ctrl_connection_unref(ctrl_conn);
1525         udev_ctrl_unref(udev_ctrl);
1526         label_finish();
1527         udev_unref(udev);
1528         log_close();
1529         return rc;
1530 }