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 {
  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 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 {
 110         return container_of(node, struct event, node);
 111 }
 112
 113 static void event_queue_cleanup(struct udev *udev, enum event_state type);
 114
 115 enum worker_state {
 116         WORKER_UNDEF,
 117         WORKER_RUNNING,
 118         WORKER_IDLE,
 119         WORKER_KILLED,
 120 };
 121
 122 struct worker {
 123         struct udev_list_node node;
 124         struct udev *udev;
 125         int refcount;
 126         pid_t pid;
 127         struct udev_monitor *monitor;
 128         enum worker_state state;
 129         struct event *event;
 130         usec_t event_start_usec;
 131 };
 132
 133 /* passed from worker to main process */
 134 struct worker_message {
 135         pid_t pid;
 136         int exitcode;
 137 };
 138
 139 static inline struct worker *node_to_worker(struct udev_list_node *node)
 140 {
 141         return container_of(node, struct worker, node);
 142 }
 143
 144 static void event_queue_delete(struct event *event)
 145 {
 146         udev_list_node_remove(&event->node);
 147         udev_device_unref(event->dev);
 148         free(event);
 149 }
 150
 151 static struct worker *worker_ref(struct worker *worker)
 152 {
 153         worker->refcount++;
 154         return worker;
 155 }
 156
 157 static void worker_cleanup(struct worker *worker)
 158 {
 159         udev_list_node_remove(&worker->node);
 160         udev_monitor_unref(worker->monitor);
 161         children--;
 162         free(worker);
 163 }
 164
 165 static void worker_unref(struct worker *worker)
 166 {
 167         worker->refcount--;
 168         if (worker->refcount > 0)
 169                 return;
 170         log_debug("worker [%u] cleaned up", worker->pid);
 171         worker_cleanup(worker);
 172 }
 173
 174 static void worker_list_cleanup(struct udev *udev)
 175 {
 176         struct udev_list_node *loop, *tmp;
 177
 178         udev_list_node_foreach_safe(loop, tmp, &worker_list) {
 179                 struct worker *worker = node_to_worker(loop);
 180
 181                 worker_cleanup(worker);
 182         }
 183 }
 184
 185 static void worker_new(struct event *event)
 186 {
 187         struct udev *udev = event->udev;
 188         struct worker *worker;
 189         struct udev_monitor *worker_monitor;
 190         pid_t pid;
 191
 192         /* listen for new events */
 193         worker_monitor = udev_monitor_new_from_netlink(udev, NULL);
 194         if (worker_monitor == NULL)
 195                 return;
 196         /* allow the main daemon netlink address to send devices to the worker */
 197         udev_monitor_allow_unicast_sender(worker_monitor, monitor);
 198         udev_monitor_enable_receiving(worker_monitor);
 199
 200         worker = new0(struct worker, 1);
 201         if (worker == NULL) {
 202                 udev_monitor_unref(worker_monitor);
 203                 return;
 204         }
 205         /* worker + event reference */
 206         worker->refcount = 2;
 207         worker->udev = udev;
 208
 209         pid = fork();
 210         switch (pid) {
 211         case 0: {
 212                 struct udev_device *dev = NULL;
 213                 int fd_monitor;
 214                 struct epoll_event ep_signal, ep_monitor;
 215                 sigset_t mask;
 216                 int rc = EXIT_SUCCESS;
 217
 218                 /* take initial device from queue */
 219                 dev = event->dev;
 220                 event->dev = NULL;
 221
 222                 free(worker);
 223                 worker_list_cleanup(udev);
 224                 event_queue_cleanup(udev, EVENT_UNDEF);
 225                 udev_monitor_unref(monitor);
 226                 udev_ctrl_unref(udev_ctrl);
 227                 close(fd_signal);
 228                 close(fd_ep);
 229                 close(worker_watch[READ_END]);
 230
 231                 sigfillset(&mask);
 232                 fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
 233                 if (fd_signal < 0) {
 234                         log_error("error creating signalfd %m");
 235                         rc = 2;
 236                         goto out;
 237                 }
 238
 239                 fd_ep = epoll_create1(EPOLL_CLOEXEC);
 240                 if (fd_ep < 0) {
 241                         log_error("error creating epoll fd: %m");
 242                         rc = 3;
 243                         goto out;
 244                 }
 245
 246                 memzero(&ep_signal, sizeof(struct epoll_event));
 247                 ep_signal.events = EPOLLIN;
 248                 ep_signal.data.fd = fd_signal;
 249
 250                 fd_monitor = udev_monitor_get_fd(worker_monitor);
 251                 memzero(&ep_monitor, sizeof(struct epoll_event));
 252                 ep_monitor.events = EPOLLIN;
 253                 ep_monitor.data.fd = fd_monitor;
 254
 255                 if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
 256                     epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_monitor, &ep_monitor) < 0) {
 257                         log_error("fail to add fds to epoll: %m");
 258                         rc = 4;
 259                         goto out;
 260                 }
 261
 262                 /* request TERM signal if parent exits */
 263                 prctl(PR_SET_PDEATHSIG, SIGTERM);
 264
 265                 /* reset OOM score, we only protect the main daemon */
 266                 write_string_file("/proc/self/oom_score_adj", "0");
 267
 268                 for (;;) {
 269                         struct udev_event *udev_event;
 270                         struct worker_message msg;
 271                         int fd_lock = -1;
 272                         int err = 0;
 273
 274                         log_debug("seq %llu running", udev_device_get_seqnum(dev));
 275                         udev_event = udev_event_new(dev);
 276                         if (udev_event == NULL) {
 277                                 rc = 5;
 278                                 goto out;
 279                         }
 280
 281                         /* needed for SIGCHLD/SIGTERM in spawn() */
 282                         udev_event->fd_signal = fd_signal;
 283
 284                         if (exec_delay > 0)
 285                                 udev_event->exec_delay = exec_delay;
 286
 287                         /*
 288                          * Take a shared lock on the device node; this establishes
 289                          * a concept of device "ownership" to serialize device
 290                          * access. External processes holding an exclusive lock will
 291                          * cause udev to skip the event handling; in the case udev
 292                          * acquired the lock, the external process can block until
 293                          * udev has finished its event handling.
 294                          */
 295                         if (!streq_ptr(udev_device_get_action(dev), "remove") &&
 296                             streq_ptr("block", udev_device_get_subsystem(dev)) &&
 297                             !startswith(udev_device_get_sysname(dev), "dm-") &&
 298                             !startswith(udev_device_get_sysname(dev), "md")) {
 299                                 struct udev_device *d = dev;
 300
 301                                 if (streq_ptr("partition", udev_device_get_devtype(d)))
 302                                         d = udev_device_get_parent(d);
 303
 304                                 if (d) {
 305                                         fd_lock = open(udev_device_get_devnode(d), O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
 306                                         if (fd_lock >= 0 && flock(fd_lock, LOCK_SH|LOCK_NB) < 0) {
 307                                                 log_debug("Unable to flock(%s), skipping event handling: %m", udev_device_get_devnode(d));
 308                                                 err = -EWOULDBLOCK;
 309                                                 fd_lock = safe_close(fd_lock);
 310                                                 goto skip;
 311                                         }
 312                                 }
 313                         }
 314
 315                         /* apply rules, create node, symlinks */
 316                         udev_event_execute_rules(udev_event, rules, &sigmask_orig);
 317
 318                         udev_event_execute_run(udev_event, &sigmask_orig);
 319
 320                         /* apply/restore inotify watch */
 321                         if (udev_event->inotify_watch) {
 322                                 udev_watch_begin(udev, dev);
 323                                 udev_device_update_db(dev);
 324                         }
 325
 326                         safe_close(fd_lock);
 327
 328                         /* send processed event back to libudev listeners */
 329                         udev_monitor_send_device(worker_monitor, NULL, dev);
 330
 331 skip:
 332                         /* send udevd the result of the event execution */
 333                         memzero(&msg, sizeof(struct worker_message));
 334                         msg.exitcode = err;
 335                         msg.pid = getpid();
 336                         send(worker_watch[WRITE_END], &msg, sizeof(struct worker_message), 0);
 337
 338                         log_debug("seq %llu processed with %i", udev_device_get_seqnum(dev), err);
 339
 340                         udev_device_unref(dev);
 341                         dev = NULL;
 342
 343                         if (udev_event->sigterm) {
 344                                 udev_event_unref(udev_event);
 345                                 goto out;
 346                         }
 347
 348                         udev_event_unref(udev_event);
 349
 350                         /* wait for more device messages from main udevd, or term signal */
 351                         while (dev == NULL) {
 352                                 struct epoll_event ev[4];
 353                                 int fdcount;
 354                                 int i;
 355
 356                                 fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), -1);
 357                                 if (fdcount < 0) {
 358                                         if (errno == EINTR)
 359                                                 continue;
 360                                         log_error("failed to poll: %m");
 361                                         goto out;
 362                                 }
 363
 364                                 for (i = 0; i < fdcount; i++) {
 365                                         if (ev[i].data.fd == fd_monitor && ev[i].events & EPOLLIN) {
 366                                                 dev = udev_monitor_receive_device(worker_monitor);
 367                                                 break;
 368                                         } else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN) {
 369                                                 struct signalfd_siginfo fdsi;
 370                                                 ssize_t size;
 371
 372                                                 size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
 373                                                 if (size != sizeof(struct signalfd_siginfo))
 374                                                         continue;
 375                                                 switch (fdsi.ssi_signo) {
 376                                                 case SIGTERM:
 377                                                         goto out;
 378                                                 }
 379                                         }
 380                                 }
 381                         }
 382                 }
 383 out:
 384                 udev_device_unref(dev);
 385                 safe_close(fd_signal);
 386                 safe_close(fd_ep);
 387                 close(fd_inotify);
 388                 close(worker_watch[WRITE_END]);
 389                 udev_rules_unref(rules);
 390                 udev_builtin_exit(udev);
 391                 udev_monitor_unref(worker_monitor);
 392                 udev_unref(udev);
 393                 log_close();
 394                 exit(rc);
 395         }
 396         case -1:
 397                 udev_monitor_unref(worker_monitor);
 398                 event->state = EVENT_QUEUED;
 399                 free(worker);
 400                 log_error("fork of child failed: %m");
 401                 break;
 402         default:
 403                 /* close monitor, but keep address around */
 404                 udev_monitor_disconnect(worker_monitor);
 405                 worker->monitor = worker_monitor;
 406                 worker->pid = pid;
 407                 worker->state = WORKER_RUNNING;
 408                 worker->event_start_usec = now(CLOCK_MONOTONIC);
 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 {
 420         struct udev_list_node *loop;
 421
 422         udev_list_node_foreach(loop, &worker_list) {
 423                 struct worker *worker = node_to_worker(loop);
 424                 ssize_t count;
 425
 426                 if (worker->state != WORKER_IDLE)
 427                         continue;
 428
 429                 count = udev_monitor_send_device(monitor, worker->monitor, event->dev);
 430                 if (count < 0) {
 431                         log_error("worker [%u] did not accept message %zi (%m), kill it", worker->pid, count);
 432                         kill(worker->pid, SIGKILL);
 433                         worker->state = WORKER_KILLED;
 434                         continue;
 435                 }
 436                 worker_ref(worker);
 437                 worker->event = event;
 438                 worker->state = WORKER_RUNNING;
 439                 worker->event_start_usec = now(CLOCK_MONOTONIC);
 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 {
 456         struct event *event;
 457
 458         event = new0(struct event, 1);
 459         if (event == NULL)
 460                 return -1;
 461
 462         event->udev = udev_device_get_udev(dev);
 463         event->dev = dev;
 464         event->seqnum = udev_device_get_seqnum(dev);
 465         event->devpath = udev_device_get_devpath(dev);
 466         event->devpath_len = strlen(event->devpath);
 467         event->devpath_old = udev_device_get_devpath_old(dev);
 468         event->devnum = udev_device_get_devnum(dev);
 469         event->is_block = streq("block", udev_device_get_subsystem(dev));
 470         event->ifindex = udev_device_get_ifindex(dev);
 471 #ifdef HAVE_FIRMWARE
 472         if (streq(udev_device_get_subsystem(dev), "firmware"))
 473                 event->nodelay = true;
 474 #endif
 475
 476         log_debug("seq %llu queued, '%s' '%s'", udev_device_get_seqnum(dev),
 477              udev_device_get_action(dev), udev_device_get_subsystem(dev));
 478
 479         event->state = EVENT_QUEUED;
 480         udev_list_node_append(&event->node, &event_list);
 481         return 0;
 482 }
 483
 484 static void worker_kill(struct udev *udev)
 485 {
 486         struct udev_list_node *loop;
 487
 488         udev_list_node_foreach(loop, &worker_list) {
 489                 struct worker *worker = node_to_worker(loop);
 490
 491                 if (worker->state == WORKER_KILLED)
 492                         continue;
 493
 494                 worker->state = WORKER_KILLED;
 495                 kill(worker->pid, SIGTERM);
 496         }
 497 }
 498
 499 /* lookup event for identical, parent, child device */
 500 static bool is_devpath_busy(struct event *event)
 501 {
 502         struct udev_list_node *loop;
 503         size_t common;
 504
 505         /* check if queue contains events we depend on */
 506         udev_list_node_foreach(loop, &event_list) {
 507                 struct event *loop_event = node_to_event(loop);
 508
 509                 /* we already found a later event, earlier can not block us, no need to check again */
 510                 if (loop_event->seqnum < event->delaying_seqnum)
 511                         continue;
 512
 513                 /* event we checked earlier still exists, no need to check again */
 514                 if (loop_event->seqnum == event->delaying_seqnum)
 515                         return true;
 516
 517                 /* found ourself, no later event can block us */
 518                 if (loop_event->seqnum >= event->seqnum)
 519                         break;
 520
 521                 /* check major/minor */
 522                 if (major(event->devnum) != 0 && event->devnum == loop_event->devnum && event->is_block == loop_event->is_block)
 523                         return true;
 524
 525                 /* check network device ifindex */
 526                 if (event->ifindex != 0 && event->ifindex == loop_event->ifindex)
 527                         return true;
 528
 529                 /* check our old name */
 530                 if (event->devpath_old != NULL && streq(loop_event->devpath, event->devpath_old)) {
 531                         event->delaying_seqnum = loop_event->seqnum;
 532                         return true;
 533                 }
 534
 535                 /* compare devpath */
 536                 common = MIN(loop_event->devpath_len, event->devpath_len);
 537
 538                 /* one devpath is contained in the other? */
 539                 if (memcmp(loop_event->devpath, event->devpath, common) != 0)
 540                         continue;
 541
 542                 /* identical device event found */
 543                 if (loop_event->devpath_len == event->devpath_len) {
 544                         /* devices names might have changed/swapped in the meantime */
 545                         if (major(event->devnum) != 0 && (event->devnum != loop_event->devnum || event->is_block != loop_event->is_block))
 546                                 continue;
 547                         if (event->ifindex != 0 && event->ifindex != loop_event->ifindex)
 548                                 continue;
 549                         event->delaying_seqnum = loop_event->seqnum;
 550                         return true;
 551                 }
 552
 553 #ifdef HAVE_FIRMWARE
 554                 /* allow to bypass the dependency tracking */
 555                 if (event->nodelay)
 556                         continue;
 557 #endif
 558
 559                 /* parent device event found */
 560                 if (event->devpath[common] == '/') {
 561                         event->delaying_seqnum = loop_event->seqnum;
 562                         return true;
 563                 }
 564
 565                 /* child device event found */
 566                 if (loop_event->devpath[common] == '/') {
 567                         event->delaying_seqnum = loop_event->seqnum;
 568                         return true;
 569                 }
 570
 571                 /* no matching device */
 572                 continue;
 573         }
 574
 575         return false;
 576 }
 577
 578 static void event_queue_start(struct udev *udev)
 579 {
 580         struct udev_list_node *loop;
 581
 582         udev_list_node_foreach(loop, &event_list) {
 583                 struct event *event = node_to_event(loop);
 584
 585                 if (event->state != EVENT_QUEUED)
 586                         continue;
 587
 588                 /* do not start event if parent or child event is still running */
 589                 if (is_devpath_busy(event))
 590                         continue;
 591
 592                 event_run(event);
 593         }
 594 }
 595
 596 static void event_queue_cleanup(struct udev *udev, enum event_state match_type)
 597 {
 598         struct udev_list_node *loop, *tmp;
 599
 600         udev_list_node_foreach_safe(loop, tmp, &event_list) {
 601                 struct event *event = node_to_event(loop);
 602
 603                 if (match_type != EVENT_UNDEF && match_type != event->state)
 604                         continue;
 605
 606                 event_queue_delete(event);
 607         }
 608 }
 609
 610 static void worker_returned(int fd_worker)
 611 {
 612         for (;;) {
 613                 struct worker_message msg;
 614                 ssize_t size;
 615                 struct udev_list_node *loop;
 616
 617                 size = recv(fd_worker, &msg, sizeof(struct worker_message), MSG_DONTWAIT);
 618                 if (size != sizeof(struct worker_message))
 619                         break;
 620
 621                 /* lookup worker who sent the signal */
 622                 udev_list_node_foreach(loop, &worker_list) {
 623                         struct worker *worker = node_to_worker(loop);
 624
 625                         if (worker->pid != msg.pid)
 626                                 continue;
 627
 628                         /* worker returned */
 629                         if (worker->event) {
 630                                 worker->event->exitcode = msg.exitcode;
 631                                 event_queue_delete(worker->event);
 632                                 worker->event = NULL;
 633                         }
 634                         if (worker->state != WORKER_KILLED)
 635                                 worker->state = WORKER_IDLE;
 636                         worker_unref(worker);
 637                         break;
 638                 }
 639         }
 640 }
 641
 642 /* receive the udevd message from userspace */
 643 static struct udev_ctrl_connection *handle_ctrl_msg(struct udev_ctrl *uctrl)
 644 {
 645         struct udev *udev = udev_ctrl_get_udev(uctrl);
 646         struct udev_ctrl_connection *ctrl_conn;
 647         struct udev_ctrl_msg *ctrl_msg = NULL;
 648         const char *str;
 649         int i;
 650
 651         ctrl_conn = udev_ctrl_get_connection(uctrl);
 652         if (ctrl_conn == NULL)
 653                 goto out;
 654
 655         ctrl_msg = udev_ctrl_receive_msg(ctrl_conn);
 656         if (ctrl_msg == NULL)
 657                 goto out;
 658
 659         i = udev_ctrl_get_set_log_level(ctrl_msg);
 660         if (i >= 0) {
 661                 log_debug("udevd message (SET_LOG_LEVEL) received, log_priority=%i", i);
 662                 log_set_max_level(i);
 663                 udev_set_log_priority(udev, i);
 664                 worker_kill(udev);
 665         }
 666
 667         if (udev_ctrl_get_stop_exec_queue(ctrl_msg) > 0) {
 668                 log_debug("udevd message (STOP_EXEC_QUEUE) received");
 669                 stop_exec_queue = true;
 670         }
 671
 672         if (udev_ctrl_get_start_exec_queue(ctrl_msg) > 0) {
 673                 log_debug("udevd message (START_EXEC_QUEUE) received");
 674                 stop_exec_queue = false;
 675         }
 676
 677         if (udev_ctrl_get_reload(ctrl_msg) > 0) {
 678                 log_debug("udevd message (RELOAD) received");
 679                 reload = true;
 680         }
 681
 682         str = udev_ctrl_get_set_env(ctrl_msg);
 683         if (str != NULL) {
 684                 char *key;
 685
 686                 key = strdup(str);
 687                 if (key != NULL) {
 688                         char *val;
 689
 690                         val = strchr(key, '=');
 691                         if (val != NULL) {
 692                                 val[0] = '\0';
 693                                 val = &val[1];
 694                                 if (val[0] == '\0') {
 695                                         log_debug("udevd message (ENV) received, unset '%s'", key);
 696                                         udev_add_property(udev, key, NULL);
 697                                 } else {
 698                                         log_debug("udevd message (ENV) received, set '%s=%s'", key, val);
 699                                         udev_add_property(udev, key, val);
 700                                 }
 701                         } else {
 702                                 log_error("wrong key format '%s'", key);
 703                         }
 704                         free(key);
 705                 }
 706                 worker_kill(udev);
 707         }
 708
 709         i = udev_ctrl_get_set_children_max(ctrl_msg);
 710         if (i >= 0) {
 711                 log_debug("udevd message (SET_MAX_CHILDREN) received, children_max=%i", i);
 712                 children_max = i;
 713         }
 714
 715         if (udev_ctrl_get_ping(ctrl_msg) > 0)
 716                 log_debug("udevd message (SYNC) received");
 717
 718         if (udev_ctrl_get_exit(ctrl_msg) > 0) {
 719                 log_debug("udevd message (EXIT) received");
 720                 udev_exit = true;
 721                 /* keep reference to block the client until we exit */
 722                 udev_ctrl_connection_ref(ctrl_conn);
 723         }
 724 out:
 725         udev_ctrl_msg_unref(ctrl_msg);
 726         return udev_ctrl_connection_unref(ctrl_conn);
 727 }
 728
 729 static int synthesize_change(struct udev_device *dev) {
 730         char filename[UTIL_PATH_SIZE];
 731         int r;
 732
 733         if (streq_ptr("block", udev_device_get_subsystem(dev)) &&
 734             streq_ptr("disk", udev_device_get_devtype(dev)) &&
 735             !startswith(udev_device_get_sysname(dev), "dm-")) {
 736                 bool part_table_read = false;
 737                 bool has_partitions = false;
 738                 int fd;
 739                 struct udev *udev = udev_device_get_udev(dev);
 740                 _cleanup_udev_enumerate_unref_ struct udev_enumerate *e = NULL;
 741                 struct udev_list_entry *item;
 742
 743                 /*
 744                  * Try to re-read the partition table. This only succeeds if
 745                  * none of the devices is busy. The kernel returns 0 if no
 746                  * partition table is found, and we will not get an event for
 747                  * the disk.
 748                  */
 749                 fd = open(udev_device_get_devnode(dev), O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
 750                 if (fd >= 0) {
 751                         r = flock(fd, LOCK_EX|LOCK_NB);
 752                         if (r >= 0)
 753                                 r = ioctl(fd, BLKRRPART, 0);
 754
 755                         close(fd);
 756                         if (r >= 0)
 757                                 part_table_read = true;
 758                 }
 759
 760                 /* search for partitions */
 761                 e = udev_enumerate_new(udev);
 762                 if (!e)
 763                         return -ENOMEM;
 764
 765                 r = udev_enumerate_add_match_parent(e, dev);
 766                 if (r < 0)
 767                         return r;
 768
 769                 r = udev_enumerate_add_match_subsystem(e, "block");
 770                 if (r < 0)
 771                         return r;
 772
 773                 r = udev_enumerate_scan_devices(e);
 774                 if (r < 0)
 775                         return r;
 776
 777                 udev_list_entry_foreach(item, udev_enumerate_get_list_entry(e)) {
 778                         _cleanup_udev_device_unref_ struct udev_device *d = NULL;
 779
 780                         d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
 781                         if (!d)
 782                                 continue;
 783
 784                         if (!streq_ptr("partition", udev_device_get_devtype(d)))
 785                                 continue;
 786
 787                         has_partitions = true;
 788                         break;
 789                 }
 790
 791                 /*
 792                  * We have partitions and re-read the table, the kernel already sent
 793                  * out a "change" event for the disk, and "remove/add" for all
 794                  * partitions.
 795                  */
 796                 if (part_table_read && has_partitions)
 797                         return 0;
 798
 799                 /*
 800                  * We have partitions but re-reading the partition table did not
 801                  * work, synthesize "change" for the disk and all partitions.
 802                  */
 803                 log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev));
 804                 strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
 805                 write_string_file(filename, "change");
 806
 807                 udev_list_entry_foreach(item, udev_enumerate_get_list_entry(e)) {
 808                         _cleanup_udev_device_unref_ struct udev_device *d = NULL;
 809
 810                         d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
 811                         if (!d)
 812                                 continue;
 813
 814                         if (!streq_ptr("partition", udev_device_get_devtype(d)))
 815                                 continue;
 816
 817                         log_debug("device %s closed, synthesising partition '%s' 'change'",
 818                                   udev_device_get_devnode(dev), udev_device_get_devnode(d));
 819                         strscpyl(filename, sizeof(filename), udev_device_get_syspath(d), "/uevent", NULL);
 820                         write_string_file(filename, "change");
 821                 }
 822
 823                 return 0;
 824         }
 825
 826         log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev));
 827         strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
 828         write_string_file(filename, "change");
 829
 830         return 0;
 831 }
 832
 833 static int handle_inotify(struct udev *udev)
 834 {
 835         int nbytes, pos;
 836         char *buf;
 837         struct inotify_event *ev;
 838         int r;
 839
 840         r = ioctl(fd_inotify, FIONREAD, &nbytes);
 841         if (r < 0 || nbytes <= 0)
 842                 return -errno;
 843
 844         buf = malloc(nbytes);
 845         if (!buf) {
 846                 log_error("error getting buffer for inotify");
 847                 return -ENOMEM;
 848         }
 849
 850         nbytes = read(fd_inotify, buf, nbytes);
 851
 852         for (pos = 0; pos < nbytes; pos += sizeof(struct inotify_event) + ev->len) {
 853                 struct udev_device *dev;
 854
 855                 ev = (struct inotify_event *)(buf + pos);
 856                 dev = udev_watch_lookup(udev, ev->wd);
 857                 if (!dev)
 858                         continue;
 859
 860                 log_debug("inotify event: %x for %s", ev->mask, udev_device_get_devnode(dev));
 861                 if (ev->mask & IN_CLOSE_WRITE)
 862                         synthesize_change(dev);
 863                 else if (ev->mask & IN_IGNORED)
 864                         udev_watch_end(udev, dev);
 865
 866                 udev_device_unref(dev);
 867         }
 868
 869         free(buf);
 870         return 0;
 871 }
 872
 873 static void handle_signal(struct udev *udev, int signo)
 874 {
 875         switch (signo) {
 876         case SIGINT:
 877         case SIGTERM:
 878                 udev_exit = true;
 879                 break;
 880         case SIGCHLD:
 881                 for (;;) {
 882                         pid_t pid;
 883                         int status;
 884                         struct udev_list_node *loop, *tmp;
 885
 886                         pid = waitpid(-1, &status, WNOHANG);
 887                         if (pid <= 0)
 888                                 break;
 889
 890                         udev_list_node_foreach_safe(loop, tmp, &worker_list) {
 891                                 struct worker *worker = node_to_worker(loop);
 892
 893                                 if (worker->pid != pid)
 894                                         continue;
 895                                 log_debug("worker [%u] exit", pid);
 896
 897                                 if (WIFEXITED(status)) {
 898                                         if (WEXITSTATUS(status) != 0)
 899                                                 log_error("worker [%u] exit with return code %i",
 900                                                           pid, WEXITSTATUS(status));
 901                                 } else if (WIFSIGNALED(status)) {
 902                                         log_error("worker [%u] terminated by signal %i (%s)",
 903                                                   pid, WTERMSIG(status), strsignal(WTERMSIG(status)));
 904                                 } else if (WIFSTOPPED(status)) {
 905                                         log_error("worker [%u] stopped", pid);
 906                                 } else if (WIFCONTINUED(status)) {
 907                                         log_error("worker [%u] continued", pid);
 908                                 } else {
 909                                         log_error("worker [%u] exit with status 0x%04x", pid, status);
 910                                 }
 911
 912                                 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 913                                         if (worker->event) {
 914                                                 log_error("worker [%u] failed while handling '%s'",
 915                                                           pid, worker->event->devpath);
 916                                                 worker->event->exitcode = -32;
 917                                                 event_queue_delete(worker->event);
 918
 919                                                 /* drop reference taken for state 'running' */
 920                                                 worker_unref(worker);
 921                                         }
 922                                 }
 923                                 worker_unref(worker);
 924                                 break;
 925                         }
 926                 }
 927                 break;
 928         case SIGHUP:
 929                 reload = true;
 930                 break;
 931         }
 932 }
 933
 934 static int systemd_fds(struct udev *udev, int *rctrl, int *rnetlink)
 935 {
 936         int ctrl = -1, netlink = -1;
 937         int fd, n;
 938
 939         n = sd_listen_fds(true);
 940         if (n <= 0)
 941                 return -1;
 942
 943         for (fd = SD_LISTEN_FDS_START; fd < n + SD_LISTEN_FDS_START; fd++) {
 944                 if (sd_is_socket(fd, AF_LOCAL, SOCK_SEQPACKET, -1)) {
 945                         if (ctrl >= 0)
 946                                 return -1;
 947                         ctrl = fd;
 948                         continue;
 949                 }
 950
 951                 if (sd_is_socket(fd, AF_NETLINK, SOCK_RAW, -1)) {
 952                         if (netlink >= 0)
 953                                 return -1;
 954                         netlink = fd;
 955                         continue;
 956                 }
 957
 958                 return -1;
 959         }
 960
 961         if (ctrl < 0 || netlink < 0)
 962                 return -1;
 963
 964         log_debug("ctrl=%i netlink=%i", ctrl, netlink);
 965         *rctrl = ctrl;
 966         *rnetlink = netlink;
 967         return 0;
 968 }
 969
 970 /*
 971  * read the kernel commandline, in case we need to get into debug mode
 972  *   udev.log-priority=<level>              syslog priority
 973  *   udev.children-max=<number of workers>  events are fully serialized if set to 1
 974  *   udev.exec-delay=<number of seconds>    delay execution of every executed program
 975  */
 976 static void kernel_cmdline_options(struct udev *udev)
 977 {
 978         _cleanup_free_ char *line = NULL;
 979         char *w, *state;
 980         size_t l;
 981         int r;
 982
 983         r = proc_cmdline(&line);
 984         if (r < 0)
 985                 log_warning("Failed to read /proc/cmdline, ignoring: %s", strerror(-r));
 986         if (r <= 0)
 987                 return;
 988
 989         FOREACH_WORD_QUOTED(w, l, line, state) {
 990                 char *s, *opt;
 991
 992                 s = strndup(w, l);
 993                 if (!s)
 994                         break;
 995
 996                 /* accept the same options for the initrd, prefixed with "rd." */
 997                 if (in_initrd() && startswith(s, "rd."))
 998                         opt = s + 3;
 999                 else
1000                         opt = s;
1001
1002                 if (startswith(opt, "udev.log-priority=")) {
1003                         int prio;
1004
1005                         prio = util_log_priority(opt + 18);
1006                         log_set_max_level(prio);
1007                         udev_set_log_priority(udev, prio);
1008                 } else if (startswith(opt, "udev.children-max=")) {
1009                         children_max = strtoul(opt + 18, NULL, 0);
1010                 } else if (startswith(opt, "udev.exec-delay=")) {
1011                         exec_delay = strtoul(opt + 16, NULL, 0);
1012                 }
1013
1014                 free(s);
1015         }
1016 }
1017
1018 int main(int argc, char *argv[])
1019 {
1020         struct udev *udev;
1021         sigset_t mask;
1022         int daemonize = false;
1023         int resolve_names = 1;
1024         static const struct option options[] = {
1025                 { "daemon", no_argument, NULL, 'd' },
1026                 { "debug", no_argument, NULL, 'D' },
1027                 { "children-max", required_argument, NULL, 'c' },
1028                 { "exec-delay", required_argument, NULL, 'e' },
1029                 { "resolve-names", required_argument, NULL, 'N' },
1030                 { "help", no_argument, NULL, 'h' },
1031                 { "version", no_argument, NULL, 'V' },
1032                 {}
1033         };
1034         int fd_ctrl = -1;
1035         int fd_netlink = -1;
1036         int fd_worker = -1;
1037         struct epoll_event ep_ctrl, ep_inotify, ep_signal, ep_netlink, ep_worker;
1038         struct udev_ctrl_connection *ctrl_conn = NULL;
1039         int rc = 1;
1040
1041         udev = udev_new();
1042         if (udev == NULL)
1043                 goto exit;
1044
1045         log_set_target(LOG_TARGET_AUTO);
1046         log_parse_environment();
1047         log_open();
1048
1049         udev_set_log_fn(udev, udev_main_log);
1050         log_set_max_level(udev_get_log_priority(udev));
1051
1052         log_debug("version %s", VERSION);
1053         label_init("/dev");
1054
1055         for (;;) {
1056                 int option;
1057
1058                 option = getopt_long(argc, argv, "c:de:DtN:hV", options, NULL);
1059                 if (option == -1)
1060                         break;
1061
1062                 switch (option) {
1063                 case 'd':
1064                         daemonize = true;
1065                         break;
1066                 case 'c':
1067                         children_max = strtoul(optarg, NULL, 0);
1068                         break;
1069                 case 'e':
1070                         exec_delay = strtoul(optarg, NULL, 0);
1071                         break;
1072                 case 'D':
1073                         debug = true;
1074                         log_set_max_level(LOG_DEBUG);
1075                         udev_set_log_priority(udev, LOG_DEBUG);
1076                         break;
1077                 case 'N':
1078                         if (streq(optarg, "early")) {
1079                                 resolve_names = 1;
1080                         } else if (streq(optarg, "late")) {
1081                                 resolve_names = 0;
1082                         } else if (streq(optarg, "never")) {
1083                                 resolve_names = -1;
1084                         } else {
1085                                 fprintf(stderr, "resolve-names must be early, late or never\n");
1086                                 log_error("resolve-names must be early, late or never");
1087                                 goto exit;
1088                         }
1089                         break;
1090                 case 'h':
1091                         printf("Usage: udevd OPTIONS\n"
1092                                "  --daemon\n"
1093                                "  --debug\n"
1094                                "  --children-max=<maximum number of workers>\n"
1095                                "  --exec-delay=<seconds to wait before executing RUN=>\n"
1096                                "  --resolve-names=early|late|never\n"
1097                                "  --version\n"
1098                                "  --help\n"
1099                                "\n");
1100                         goto exit;
1101                 case 'V':
1102                         printf("%s\n", VERSION);
1103                         goto exit;
1104                 default:
1105                         goto exit;
1106                 }
1107         }
1108
1109         kernel_cmdline_options(udev);
1110
1111         if (getuid() != 0) {
1112                 fprintf(stderr, "root privileges required\n");
1113                 log_error("root privileges required");
1114                 goto exit;
1115         }
1116
1117         /* set umask before creating any file/directory */
1118         chdir("/");
1119         umask(022);
1120
1121         mkdir("/run/udev", 0755);
1122
1123         dev_setup(NULL);
1124
1125         /* before opening new files, make sure std{in,out,err} fds are in a sane state */
1126         if (daemonize) {
1127                 int fd;
1128
1129                 fd = open("/dev/null", O_RDWR);
1130                 if (fd >= 0) {
1131                         if (write(STDOUT_FILENO, 0, 0) < 0)
1132                                 dup2(fd, STDOUT_FILENO);
1133                         if (write(STDERR_FILENO, 0, 0) < 0)
1134                                 dup2(fd, STDERR_FILENO);
1135                         if (fd > STDERR_FILENO)
1136                                 close(fd);
1137                 } else {
1138                         fprintf(stderr, "cannot open /dev/null\n");
1139                         log_error("cannot open /dev/null");
1140                 }
1141         }
1142
1143         if (systemd_fds(udev, &fd_ctrl, &fd_netlink) >= 0) {
1144                 /* get control and netlink socket from systemd */
1145                 udev_ctrl = udev_ctrl_new_from_fd(udev, fd_ctrl);
1146                 if (udev_ctrl == NULL) {
1147                         log_error("error taking over udev control socket");
1148                         rc = 1;
1149                         goto exit;
1150                 }
1151
1152                 monitor = udev_monitor_new_from_netlink_fd(udev, "kernel", fd_netlink);
1153                 if (monitor == NULL) {
1154                         log_error("error taking over netlink socket");
1155                         rc = 3;
1156                         goto exit;
1157                 }
1158
1159                 /* get our own cgroup, we regularly kill everything udev has left behind */
1160                 if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &udev_cgroup) < 0)
1161                         udev_cgroup = NULL;
1162         } else {
1163                 /* open control and netlink socket */
1164                 udev_ctrl = udev_ctrl_new(udev);
1165                 if (udev_ctrl == NULL) {
1166                         fprintf(stderr, "error initializing udev control socket");
1167                         log_error("error initializing udev control socket");
1168                         rc = 1;
1169                         goto exit;
1170                 }
1171                 fd_ctrl = udev_ctrl_get_fd(udev_ctrl);
1172
1173                 monitor = udev_monitor_new_from_netlink(udev, "kernel");
1174                 if (monitor == NULL) {
1175                         fprintf(stderr, "error initializing netlink socket\n");
1176                         log_error("error initializing netlink socket");
1177                         rc = 3;
1178                         goto exit;
1179                 }
1180                 fd_netlink = udev_monitor_get_fd(monitor);
1181         }
1182
1183         if (udev_monitor_enable_receiving(monitor) < 0) {
1184                 fprintf(stderr, "error binding netlink socket\n");
1185                 log_error("error binding netlink socket");
1186                 rc = 3;
1187                 goto exit;
1188         }
1189
1190         if (udev_ctrl_enable_receiving(udev_ctrl) < 0) {
1191                 fprintf(stderr, "error binding udev control socket\n");
1192                 log_error("error binding udev control socket");
1193                 rc = 1;
1194                 goto exit;
1195         }
1196
1197         udev_monitor_set_receive_buffer_size(monitor, 128 * 1024 * 1024);
1198
1199         if (daemonize) {
1200                 pid_t pid;
1201
1202                 pid = fork();
1203                 switch (pid) {
1204                 case 0:
1205                         break;
1206                 case -1:
1207                         log_error("fork of daemon failed: %m");
1208                         rc = 4;
1209                         goto exit;
1210                 default:
1211                         rc = EXIT_SUCCESS;
1212                         goto exit_daemonize;
1213                 }
1214
1215                 setsid();
1216
1217                 write_string_file("/proc/self/oom_score_adj", "-1000");
1218         } else {
1219                 sd_notify(1, "READY=1");
1220         }
1221
1222         print_kmsg("starting version " VERSION "\n");
1223
1224         if (!debug) {
1225                 int fd;
1226
1227                 fd = open("/dev/null", O_RDWR);
1228                 if (fd >= 0) {
1229                         dup2(fd, STDIN_FILENO);
1230                         dup2(fd, STDOUT_FILENO);
1231                         dup2(fd, STDERR_FILENO);
1232                         close(fd);
1233                 }
1234         }
1235
1236         fd_inotify = udev_watch_init(udev);
1237         if (fd_inotify < 0) {
1238                 fprintf(stderr, "error initializing inotify\n");
1239                 log_error("error initializing inotify");
1240                 rc = 4;
1241                 goto exit;
1242         }
1243         udev_watch_restore(udev);
1244
1245         /* block and listen to all signals on signalfd */
1246         sigfillset(&mask);
1247         sigprocmask(SIG_SETMASK, &mask, &sigmask_orig);
1248         fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
1249         if (fd_signal < 0) {
1250                 fprintf(stderr, "error creating signalfd\n");
1251                 log_error("error creating signalfd");
1252                 rc = 5;
1253                 goto exit;
1254         }
1255
1256         /* unnamed socket from workers to the main daemon */
1257         if (socketpair(AF_LOCAL, SOCK_DGRAM|SOCK_CLOEXEC, 0, worker_watch) < 0) {
1258                 fprintf(stderr, "error creating socketpair\n");
1259                 log_error("error creating socketpair");
1260                 rc = 6;
1261                 goto exit;
1262         }
1263         fd_worker = worker_watch[READ_END];
1264
1265         udev_builtin_init(udev);
1266
1267         rules = udev_rules_new(udev, resolve_names);
1268         if (rules == NULL) {
1269                 log_error("error reading rules");
1270                 goto exit;
1271         }
1272
1273         memzero(&ep_ctrl, sizeof(struct epoll_event));
1274         ep_ctrl.events = EPOLLIN;
1275         ep_ctrl.data.fd = fd_ctrl;
1276
1277         memzero(&ep_inotify, sizeof(struct epoll_event));
1278         ep_inotify.events = EPOLLIN;
1279         ep_inotify.data.fd = fd_inotify;
1280
1281         memzero(&ep_signal, sizeof(struct epoll_event));
1282         ep_signal.events = EPOLLIN;
1283         ep_signal.data.fd = fd_signal;
1284
1285         memzero(&ep_netlink, sizeof(struct epoll_event));
1286         ep_netlink.events = EPOLLIN;
1287         ep_netlink.data.fd = fd_netlink;
1288
1289         memzero(&ep_worker, sizeof(struct epoll_event));
1290         ep_worker.events = EPOLLIN;
1291         ep_worker.data.fd = fd_worker;
1292
1293         fd_ep = epoll_create1(EPOLL_CLOEXEC);
1294         if (fd_ep < 0) {
1295                 log_error("error creating epoll fd: %m");
1296                 goto exit;
1297         }
1298         if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_ctrl, &ep_ctrl) < 0 ||
1299             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_inotify, &ep_inotify) < 0 ||
1300             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
1301             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_netlink, &ep_netlink) < 0 ||
1302             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_worker, &ep_worker) < 0) {
1303                 log_error("fail to add fds to epoll: %m");
1304                 goto exit;
1305         }
1306
1307         if (children_max <= 0) {
1308                 cpu_set_t cpu_set;
1309
1310                 children_max = 8;
1311
1312                 if (sched_getaffinity(0, sizeof (cpu_set), &cpu_set) == 0) {
1313                         children_max +=  CPU_COUNT(&cpu_set) * 2;
1314                 }
1315         }
1316         log_debug("set children_max to %u", children_max);
1317
1318         rc = udev_rules_apply_static_dev_perms(rules);
1319         if (rc < 0)
1320                 log_error("failed to apply permissions on static device nodes - %s", strerror(-rc));
1321
1322         udev_list_node_init(&event_list);
1323         udev_list_node_init(&worker_list);
1324
1325         for (;;) {
1326                 static usec_t last_usec;
1327                 struct epoll_event ev[8];
1328                 int fdcount;
1329                 int timeout;
1330                 bool is_worker, is_signal, is_inotify, is_netlink, is_ctrl;
1331                 int i;
1332
1333                 if (udev_exit) {
1334                         /* close sources of new events and discard buffered events */
1335                         if (fd_ctrl >= 0) {
1336                                 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_ctrl, NULL);
1337                                 fd_ctrl = -1;
1338                         }
1339                         if (monitor != NULL) {
1340                                 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_netlink, NULL);
1341                                 udev_monitor_unref(monitor);
1342                                 monitor = NULL;
1343                         }
1344                         if (fd_inotify >= 0) {
1345                                 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_inotify, NULL);
1346                                 close(fd_inotify);
1347                                 fd_inotify = -1;
1348                         }
1349
1350                         /* discard queued events and kill workers */
1351                         event_queue_cleanup(udev, EVENT_QUEUED);
1352                         worker_kill(udev);
1353
1354                         /* exit after all has cleaned up */
1355                         if (udev_list_node_is_empty(&event_list) && children == 0)
1356                                 break;
1357
1358                         /* timeout at exit for workers to finish */
1359                         timeout = 30 * MSEC_PER_SEC;
1360                 } else if (udev_list_node_is_empty(&event_list) && children == 0) {
1361                         /* we are idle */
1362                         timeout = -1;
1363
1364                         /* cleanup possible left-over processes in our cgroup */
1365                         if (udev_cgroup)
1366                                 cg_kill(SYSTEMD_CGROUP_CONTROLLER, udev_cgroup, SIGKILL, false, true, NULL);
1367                 } else {
1368                         /* kill idle or hanging workers */
1369                         timeout = 3 * MSEC_PER_SEC;
1370                 }
1371
1372                 /* tell settle that we are busy or idle */
1373                 if (!udev_list_node_is_empty(&event_list)) {
1374                         int fd;
1375
1376                         fd = open("/run/udev/queue", O_WRONLY|O_CREAT|O_CLOEXEC|O_TRUNC|O_NOFOLLOW, 0444);
1377                         if (fd >= 0)
1378                                 close(fd);
1379                 } else {
1380                         unlink("/run/udev/queue");
1381                 }
1382
1383                 fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), timeout);
1384                 if (fdcount < 0)
1385                         continue;
1386
1387                 if (fdcount == 0) {
1388                         struct udev_list_node *loop;
1389
1390                         /* timeout */
1391                         if (udev_exit) {
1392                                 log_error("timeout, giving up waiting for workers to finish");
1393                                 break;
1394                         }
1395
1396                         /* kill idle workers */
1397                         if (udev_list_node_is_empty(&event_list)) {
1398                                 log_debug("cleanup idle workers");
1399                                 worker_kill(udev);
1400                         }
1401
1402                         /* check for hanging events */
1403                         udev_list_node_foreach(loop, &worker_list) {
1404                                 struct worker *worker = node_to_worker(loop);
1405
1406                                 if (worker->state != WORKER_RUNNING)
1407                                         continue;
1408
1409                                 if ((now(CLOCK_MONOTONIC) - worker->event_start_usec) > 30 * USEC_PER_SEC) {
1410                                         log_error("worker [%u] %s timeout; kill it", worker->pid,
1411                                             worker->event ? worker->event->devpath : "<idle>");
1412                                         kill(worker->pid, SIGKILL);
1413                                         worker->state = WORKER_KILLED;
1414
1415                                         /* drop reference taken for state 'running' */
1416                                         worker_unref(worker);
1417                                         if (worker->event) {
1418                                                 log_error("seq %llu '%s' killed", udev_device_get_seqnum(worker->event->dev), worker->event->devpath);
1419                                                 worker->event->exitcode = -64;
1420                                                 event_queue_delete(worker->event);
1421                                                 worker->event = NULL;
1422                                         }
1423                                 }
1424                         }
1425
1426                 }
1427
1428                 is_worker = is_signal = is_inotify = is_netlink = is_ctrl = false;
1429                 for (i = 0; i < fdcount; i++) {
1430                         if (ev[i].data.fd == fd_worker && ev[i].events & EPOLLIN)
1431                                 is_worker = true;
1432                         else if (ev[i].data.fd == fd_netlink && ev[i].events & EPOLLIN)
1433                                 is_netlink = true;
1434                         else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN)
1435                                 is_signal = true;
1436                         else if (ev[i].data.fd == fd_inotify && ev[i].events & EPOLLIN)
1437                                 is_inotify = true;
1438                         else if (ev[i].data.fd == fd_ctrl && ev[i].events & EPOLLIN)
1439                                 is_ctrl = true;
1440                 }
1441
1442                 /* check for changed config, every 3 seconds at most */
1443                 if ((now(CLOCK_MONOTONIC) - last_usec) > 3 * USEC_PER_SEC) {
1444                         if (udev_rules_check_timestamp(rules))
1445                                 reload = true;
1446                         if (udev_builtin_validate(udev))
1447                                 reload = true;
1448
1449                         last_usec = now(CLOCK_MONOTONIC);
1450                 }
1451
1452                 /* reload requested, HUP signal received, rules changed, builtin changed */
1453                 if (reload) {
1454                         worker_kill(udev);
1455                         rules = udev_rules_unref(rules);
1456                         udev_builtin_exit(udev);
1457                         reload = false;
1458                 }
1459
1460                 /* event has finished */
1461                 if (is_worker)
1462                         worker_returned(fd_worker);
1463
1464                 if (is_netlink) {
1465                         struct udev_device *dev;
1466
1467                         dev = udev_monitor_receive_device(monitor);
1468                         if (dev != NULL) {
1469                                 udev_device_set_usec_initialized(dev, now(CLOCK_MONOTONIC));
1470                                 if (event_queue_insert(dev) < 0)
1471                                         udev_device_unref(dev);
1472                         }
1473                 }
1474
1475                 /* start new events */
1476                 if (!udev_list_node_is_empty(&event_list) && !udev_exit && !stop_exec_queue) {
1477                         udev_builtin_init(udev);
1478                         if (rules == NULL)
1479                                 rules = udev_rules_new(udev, resolve_names);
1480                         if (rules != NULL)
1481                                 event_queue_start(udev);
1482                 }
1483
1484                 if (is_signal) {
1485                         struct signalfd_siginfo fdsi;
1486                         ssize_t size;
1487
1488                         size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
1489                         if (size == sizeof(struct signalfd_siginfo))
1490                                 handle_signal(udev, fdsi.ssi_signo);
1491                 }
1492
1493                 /* we are shutting down, the events below are not handled anymore */
1494                 if (udev_exit)
1495                         continue;
1496
1497                 /* device node watch */
1498                 if (is_inotify)
1499                         handle_inotify(udev);
1500
1501                 /*
1502                  * This needs to be after the inotify handling, to make sure,
1503                  * that the ping is send back after the possibly generated
1504                  * "change" events by the inotify device node watch.
1505                  *
1506                  * A single time we may receive a client connection which we need to
1507                  * keep open to block the client. It will be closed right before we
1508                  * exit.
1509                  */
1510                 if (is_ctrl)
1511                         ctrl_conn = handle_ctrl_msg(udev_ctrl);
1512         }
1513
1514         rc = EXIT_SUCCESS;
1515 exit:
1516         udev_ctrl_cleanup(udev_ctrl);
1517         unlink("/run/udev/queue");
1518 exit_daemonize:
1519         if (fd_ep >= 0)
1520                 close(fd_ep);
1521         worker_list_cleanup(udev);
1522         event_queue_cleanup(udev, EVENT_UNDEF);
1523         udev_rules_unref(rules);
1524         udev_builtin_exit(udev);
1525         if (fd_signal >= 0)
1526                 close(fd_signal);
1527         if (worker_watch[READ_END] >= 0)
1528                 close(worker_watch[READ_END]);
1529         if (worker_watch[WRITE_END] >= 0)
1530                 close(worker_watch[WRITE_END]);
1531         udev_monitor_unref(monitor);
1532         udev_ctrl_connection_unref(ctrl_conn);
1533         udev_ctrl_unref(udev_ctrl);
1534         label_finish();
1535         udev_unref(udev);
1536         log_close();
1537         return rc;
1538 }