udev/udevd.c

   1 /*
   2  * Copyright (C) 2004-2011 Kay Sievers <kay.sievers@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/time.h>
  35 #include <sys/prctl.h>
  36 #include <sys/socket.h>
  37 #include <sys/un.h>
  38 #include <sys/signalfd.h>
  39 #include <sys/epoll.h>
  40 #include <sys/poll.h>
  41 #include <sys/wait.h>
  42 #include <sys/stat.h>
  43 #include <sys/ioctl.h>
  44 #include <sys/inotify.h>
  45 #include <sys/utsname.h>
  46
  47 #include "udev.h"
  48 #include "sd-daemon.h"
  49
  50 static bool debug;
  51
  52 static void log_fn(struct udev *udev, int priority,
  53                    const char *file, int line, const char *fn,
  54                    const char *format, va_list args)
  55 {
  56         if (debug) {
  57                 char buf[1024];
  58                 struct timespec ts;
  59
  60                 vsnprintf(buf, sizeof(buf), format, args);
  61                 clock_gettime(CLOCK_MONOTONIC, &ts);
  62                 fprintf(stderr, "[%llu.%06u] [%u] %s: %s",
  63                         (unsigned long long) ts.tv_sec, (unsigned int) ts.tv_nsec/1000,
  64                         (int) getpid(), fn, buf);
  65         } else {
  66                 vsyslog(priority, format, args);
  67         }
  68 }
  69
  70 static struct udev_rules *rules;
  71 static struct udev_queue_export *udev_queue_export;
  72 static struct udev_ctrl *udev_ctrl;
  73 static struct udev_monitor *monitor;
  74 static int worker_watch[2] = { -1, -1 };
  75 static int fd_signal = -1;
  76 static int fd_ep = -1;
  77 static int fd_inotify = -1;
  78 static bool stop_exec_queue;
  79 static bool reload_config;
  80 static int children;
  81 static int children_max;
  82 static int exec_delay;
  83 static sigset_t sigmask_orig;
  84 static UDEV_LIST(event_list);
  85 static UDEV_LIST(worker_list);
  86 static bool udev_exit;
  87
  88 enum event_state {
  89         EVENT_UNDEF,
  90         EVENT_QUEUED,
  91         EVENT_RUNNING,
  92 };
  93
  94 struct event {
  95         struct udev_list_node node;
  96         struct udev *udev;
  97         struct udev_device *dev;
  98         enum event_state state;
  99         int exitcode;
 100         unsigned long long int delaying_seqnum;
 101         unsigned long long int seqnum;
 102         const char *devpath;
 103         size_t devpath_len;
 104         const char *devpath_old;
 105         dev_t devnum;
 106         bool is_block;
 107         int ifindex;
 108 };
 109
 110 static struct event *node_to_event(struct udev_list_node *node)
 111 {
 112         char *event;
 113
 114         event = (char *)node;
 115         event -= offsetof(struct event, node);
 116         return (struct event *)event;
 117 }
 118
 119 static void event_queue_cleanup(struct udev *udev, enum event_state type);
 120
 121 enum worker_state {
 122         WORKER_UNDEF,
 123         WORKER_RUNNING,
 124         WORKER_IDLE,
 125         WORKER_KILLED,
 126 };
 127
 128 struct worker {
 129         struct udev_list_node node;
 130         struct udev *udev;
 131         int refcount;
 132         pid_t pid;
 133         struct udev_monitor *monitor;
 134         enum worker_state state;
 135         struct event *event;
 136 };
 137
 138 /* passed from worker to main process */
 139 struct worker_message {
 140         pid_t pid;
 141         int exitcode;
 142 };
 143
 144 static struct worker *node_to_worker(struct udev_list_node *node)
 145 {
 146         char *worker;
 147
 148         worker = (char *)node;
 149         worker -= offsetof(struct worker, node);
 150         return (struct worker *)worker;
 151 }
 152
 153 static void event_queue_delete(struct event *event, bool export)
 154 {
 155         udev_list_node_remove(&event->node);
 156
 157         if (export) {
 158                 udev_queue_export_device_finished(udev_queue_export, event->dev);
 159                 info(event->udev, "seq %llu done with %i\n", udev_device_get_seqnum(event->dev), event->exitcode);
 160         }
 161         udev_device_unref(event->dev);
 162         free(event);
 163 }
 164
 165 static struct worker *worker_ref(struct worker *worker)
 166 {
 167         worker->refcount++;
 168         return worker;
 169 }
 170
 171 static void worker_cleanup(struct worker *worker)
 172 {
 173         udev_list_node_remove(&worker->node);
 174         udev_monitor_unref(worker->monitor);
 175         children--;
 176         free(worker);
 177 }
 178
 179 static void worker_unref(struct worker *worker)
 180 {
 181         worker->refcount--;
 182         if (worker->refcount > 0)
 183                 return;
 184         info(worker->udev, "worker [%u] cleaned up\n", worker->pid);
 185         worker_cleanup(worker);
 186 }
 187
 188 static void worker_list_cleanup(struct udev *udev)
 189 {
 190         struct udev_list_node *loop, *tmp;
 191
 192         udev_list_node_foreach_safe(loop, tmp, &worker_list) {
 193                 struct worker *worker = node_to_worker(loop);
 194
 195                 worker_cleanup(worker);
 196         }
 197 }
 198
 199 static void worker_new(struct event *event)
 200 {
 201         struct udev *udev = event->udev;
 202         struct worker *worker;
 203         struct udev_monitor *worker_monitor;
 204         pid_t pid;
 205
 206         /* listen for new events */
 207         worker_monitor = udev_monitor_new_from_netlink(udev, NULL);
 208         if (worker_monitor == NULL)
 209                 return;
 210         /* allow the main daemon netlink address to send devices to the worker */
 211         udev_monitor_allow_unicast_sender(worker_monitor, monitor);
 212         udev_monitor_enable_receiving(worker_monitor);
 213
 214         worker = calloc(1, sizeof(struct worker));
 215         if (worker == NULL) {
 216                 udev_monitor_unref(worker_monitor);
 217                 return;
 218         }
 219         /* worker + event reference */
 220         worker->refcount = 2;
 221         worker->udev = udev;
 222
 223         pid = fork();
 224         switch (pid) {
 225         case 0: {
 226                 struct udev_device *dev = NULL;
 227                 int fd_monitor;
 228                 struct epoll_event ep_signal, ep_monitor;
 229                 sigset_t mask;
 230                 int rc = EXIT_SUCCESS;
 231
 232                 /* move initial device from queue */
 233                 dev = event->dev;
 234                 event->dev = NULL;
 235
 236                 free(worker);
 237                 worker_list_cleanup(udev);
 238                 event_queue_cleanup(udev, EVENT_UNDEF);
 239                 udev_queue_export_unref(udev_queue_export);
 240                 udev_monitor_unref(monitor);
 241                 udev_ctrl_unref(udev_ctrl);
 242                 close(fd_signal);
 243                 close(fd_ep);
 244                 close(worker_watch[READ_END]);
 245
 246                 sigfillset(&mask);
 247                 fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
 248                 if (fd_signal < 0) {
 249                         err(udev, "error creating signalfd %m\n");
 250                         rc = 2;
 251                         goto out;
 252                 }
 253
 254                 fd_ep = epoll_create1(EPOLL_CLOEXEC);
 255                 if (fd_ep < 0) {
 256                         err(udev, "error creating epoll fd: %m\n");
 257                         rc = 3;
 258                         goto out;
 259                 }
 260
 261                 memset(&ep_signal, 0, sizeof(struct epoll_event));
 262                 ep_signal.events = EPOLLIN;
 263                 ep_signal.data.fd = fd_signal;
 264
 265                 fd_monitor = udev_monitor_get_fd(worker_monitor);
 266                 memset(&ep_monitor, 0, sizeof(struct epoll_event));
 267                 ep_monitor.events = EPOLLIN;
 268                 ep_monitor.data.fd = fd_monitor;
 269
 270                 if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
 271                     epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_monitor, &ep_monitor) < 0) {
 272                         err(udev, "fail to add fds to epoll: %m\n");
 273                         rc = 4;
 274                         goto out;
 275                 }
 276
 277                 /* request TERM signal if parent exits */
 278                 prctl(PR_SET_PDEATHSIG, SIGTERM);
 279
 280                 for (;;) {
 281                         struct udev_event *udev_event;
 282                         struct worker_message msg;
 283                         int err;
 284
 285                         info(udev, "seq %llu running\n", udev_device_get_seqnum(dev));
 286                         udev_event = udev_event_new(dev);
 287                         if (udev_event == NULL) {
 288                                 rc = 5;
 289                                 goto out;
 290                         }
 291
 292                         /* needed for SIGCHLD/SIGTERM in spawn() */
 293                         udev_event->fd_signal = fd_signal;
 294
 295                         if (exec_delay > 0)
 296                                 udev_event->exec_delay = exec_delay;
 297
 298                         /* apply rules, create node, symlinks */
 299                         err = udev_event_execute_rules(udev_event, rules, &sigmask_orig);
 300
 301                         if (err == 0)
 302                                 udev_event_execute_run(udev_event, &sigmask_orig);
 303
 304                         /* apply/restore inotify watch */
 305                         if (err == 0 && udev_event->inotify_watch) {
 306                                 udev_watch_begin(udev, dev);
 307                                 udev_device_update_db(dev);
 308                         }
 309
 310                         /* send processed event back to libudev listeners */
 311                         udev_monitor_send_device(worker_monitor, NULL, dev);
 312
 313                         /* send udevd the result of the event execution */
 314                         memset(&msg, 0, sizeof(struct worker_message));
 315                         if (err != 0)
 316                                 msg.exitcode = err;
 317                         msg.pid = getpid();
 318                         send(worker_watch[WRITE_END], &msg, sizeof(struct worker_message), 0);
 319
 320                         info(udev, "seq %llu processed with %i\n", udev_device_get_seqnum(dev), err);
 321
 322                         udev_device_unref(dev);
 323                         dev = NULL;
 324
 325                         if (udev_event->sigterm) {
 326                                 udev_event_unref(udev_event);
 327                                 goto out;
 328                         }
 329
 330                         udev_event_unref(udev_event);
 331
 332                         /* wait for more device messages from main udevd, or term signal */
 333                         while (dev == NULL) {
 334                                 struct epoll_event ev[4];
 335                                 int fdcount;
 336                                 int i;
 337
 338                                 fdcount = epoll_wait(fd_ep, ev, ARRAY_SIZE(ev), -1);
 339                                 if (fdcount < 0) {
 340                                         if (errno == EINTR)
 341                                                 continue;
 342                                         err = -errno;
 343                                         err(udev, "failed to poll: %m\n");
 344                                         goto out;
 345                                 }
 346
 347                                 for (i = 0; i < fdcount; i++) {
 348                                         if (ev[i].data.fd == fd_monitor && ev[i].events & EPOLLIN) {
 349                                                 dev = udev_monitor_receive_device(worker_monitor);
 350                                         } else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN) {
 351                                                 struct signalfd_siginfo fdsi;
 352                                                 ssize_t size;
 353
 354                                                 size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
 355                                                 if (size != sizeof(struct signalfd_siginfo))
 356                                                         continue;
 357                                                 switch (fdsi.ssi_signo) {
 358                                                 case SIGTERM:
 359                                                         goto out;
 360                                                 }
 361                                         }
 362                                 }
 363                         }
 364                 }
 365 out:
 366                 udev_device_unref(dev);
 367                 if (fd_signal >= 0)
 368                         close(fd_signal);
 369                 if (fd_ep >= 0)
 370                         close(fd_ep);
 371                 close(fd_inotify);
 372                 close(worker_watch[WRITE_END]);
 373                 udev_rules_unref(rules);
 374                 udev_monitor_unref(worker_monitor);
 375                 udev_unref(udev);
 376                 udev_log_close();
 377                 exit(rc);
 378         }
 379         case -1:
 380                 udev_monitor_unref(worker_monitor);
 381                 event->state = EVENT_QUEUED;
 382                 free(worker);
 383                 err(udev, "fork of child failed: %m\n");
 384                 break;
 385         default:
 386                 /* close monitor, but keep address around */
 387                 udev_monitor_disconnect(worker_monitor);
 388                 worker->monitor = worker_monitor;
 389                 worker->pid = pid;
 390                 worker->state = WORKER_RUNNING;
 391                 worker->event = event;
 392                 event->state = EVENT_RUNNING;
 393                 udev_list_node_append(&worker->node, &worker_list);
 394                 children++;
 395                 info(udev, "seq %llu forked new worker [%u]\n", udev_device_get_seqnum(event->dev), pid);
 396                 break;
 397         }
 398 }
 399
 400 static void event_run(struct event *event, bool force)
 401 {
 402         struct udev_list_node *loop;
 403
 404         udev_list_node_foreach(loop, &worker_list) {
 405                 struct worker *worker = node_to_worker(loop);
 406                 ssize_t count;
 407
 408                 if (worker->state != WORKER_IDLE)
 409                         continue;
 410
 411                 count = udev_monitor_send_device(monitor, worker->monitor, event->dev);
 412                 if (count < 0) {
 413                         err(event->udev, "worker [%u] did not accept message %zi (%m), kill it\n", worker->pid, count);
 414                         kill(worker->pid, SIGKILL);
 415                         worker->state = WORKER_KILLED;
 416                         continue;
 417                 }
 418                 worker_ref(worker);
 419                 worker->event = event;
 420                 worker->state = WORKER_RUNNING;
 421                 event->state = EVENT_RUNNING;
 422                 return;
 423         }
 424
 425         if (!force && children >= children_max) {
 426                 if (children_max > 1)
 427                         info(event->udev, "maximum number (%i) of children reached\n", children);
 428                 return;
 429         }
 430
 431         /* start new worker and pass initial device */
 432         worker_new(event);
 433 }
 434
 435 static int event_queue_insert(struct udev_device *dev)
 436 {
 437         struct event *event;
 438
 439         event = calloc(1, sizeof(struct event));
 440         if (event == NULL)
 441                 return -1;
 442
 443         event->udev = udev_device_get_udev(dev);
 444         event->dev = dev;
 445         event->seqnum = udev_device_get_seqnum(dev);
 446         event->devpath = udev_device_get_devpath(dev);
 447         event->devpath_len = strlen(event->devpath);
 448         event->devpath_old = udev_device_get_devpath_old(dev);
 449         event->devnum = udev_device_get_devnum(dev);
 450         event->is_block = (strcmp("block", udev_device_get_subsystem(dev)) == 0);
 451         event->ifindex = udev_device_get_ifindex(dev);
 452
 453         udev_queue_export_device_queued(udev_queue_export, dev);
 454         info(event->udev, "seq %llu queued, '%s' '%s'\n", udev_device_get_seqnum(dev),
 455              udev_device_get_action(dev), udev_device_get_subsystem(dev));
 456
 457         event->state = EVENT_QUEUED;
 458         udev_list_node_append(&event->node, &event_list);
 459
 460         /* run all events with a timeout set immediately */
 461         if (udev_device_get_timeout(dev) > 0) {
 462                 event_run(event, true);
 463                 return 0;
 464         }
 465
 466         return 0;
 467 }
 468
 469 static void worker_kill(struct udev *udev, int retain)
 470 {
 471         struct udev_list_node *loop;
 472         int max;
 473
 474         if (children <= retain)
 475                 return;
 476
 477         max = children - retain;
 478
 479         udev_list_node_foreach(loop, &worker_list) {
 480                 struct worker *worker = node_to_worker(loop);
 481
 482                 if (max-- <= 0)
 483                         break;
 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 {
 496         struct udev_list_node *loop;
 497         size_t common;
 498
 499         /* check if queue contains events we depend on */
 500         udev_list_node_foreach(loop, &event_list) {
 501                 struct event *loop_event = node_to_event(loop);
 502
 503                 /* we already found a later event, earlier can not block us, no need to check again */
 504                 if (loop_event->seqnum < event->delaying_seqnum)
 505                         continue;
 506
 507                 /* event we checked earlier still exists, no need to check again */
 508                 if (loop_event->seqnum == event->delaying_seqnum)
 509                         return true;
 510
 511                 /* found ourself, no later event can block us */
 512                 if (loop_event->seqnum >= event->seqnum)
 513                         break;
 514
 515                 /* check major/minor */
 516                 if (major(event->devnum) != 0 && event->devnum == loop_event->devnum && event->is_block == loop_event->is_block)
 517                         return true;
 518
 519                 /* check network device ifindex */
 520                 if (event->ifindex != 0 && event->ifindex == loop_event->ifindex)
 521                         return true;
 522
 523                 /* check our old name */
 524                 if (event->devpath_old != NULL && strcmp(loop_event->devpath, event->devpath_old) == 0) {
 525                         event->delaying_seqnum = loop_event->seqnum;
 526                         return true;
 527                 }
 528
 529                 /* compare devpath */
 530                 common = MIN(loop_event->devpath_len, event->devpath_len);
 531
 532                 /* one devpath is contained in the other? */
 533                 if (memcmp(loop_event->devpath, event->devpath, common) != 0)
 534                         continue;
 535
 536                 /* identical device event found */
 537                 if (loop_event->devpath_len == event->devpath_len) {
 538                         /* devices names might have changed/swapped in the meantime */
 539                         if (major(event->devnum) != 0 && (event->devnum != loop_event->devnum || event->is_block != loop_event->is_block))
 540                                 continue;
 541                         if (event->ifindex != 0 && event->ifindex != loop_event->ifindex)
 542                                 continue;
 543                         event->delaying_seqnum = loop_event->seqnum;
 544                         return true;
 545                 }
 546
 547                 /* parent device event found */
 548                 if (event->devpath[common] == '/') {
 549                         event->delaying_seqnum = loop_event->seqnum;
 550                         return true;
 551                 }
 552
 553                 /* child device event found */
 554                 if (loop_event->devpath[common] == '/') {
 555                         event->delaying_seqnum = loop_event->seqnum;
 556                         return true;
 557                 }
 558
 559                 /* no matching device */
 560                 continue;
 561         }
 562
 563         return false;
 564 }
 565
 566 static void event_queue_start(struct udev *udev)
 567 {
 568         struct udev_list_node *loop;
 569
 570         udev_list_node_foreach(loop, &event_list) {
 571                 struct event *event = node_to_event(loop);
 572
 573                 if (event->state != EVENT_QUEUED)
 574                         continue;
 575
 576                 /* do not start event if parent or child event is still running */
 577                 if (is_devpath_busy(event)) {
 578                         dbg(udev, "delay seq %llu (%s)\n", event->seqnum, event->devpath);
 579                         continue;
 580                 }
 581
 582                 event_run(event, false);
 583         }
 584 }
 585
 586 static void event_queue_cleanup(struct udev *udev, enum event_state match_type)
 587 {
 588         struct udev_list_node *loop, *tmp;
 589
 590         udev_list_node_foreach_safe(loop, tmp, &event_list) {
 591                 struct event *event = node_to_event(loop);
 592
 593                 if (match_type != EVENT_UNDEF && match_type != event->state)
 594                         continue;
 595
 596                 event_queue_delete(event, false);
 597         }
 598 }
 599
 600 static void worker_returned(int fd_worker)
 601 {
 602         for (;;) {
 603                 struct worker_message msg;
 604                 ssize_t size;
 605                 struct udev_list_node *loop;
 606
 607                 size = recv(fd_worker, &msg, sizeof(struct worker_message), MSG_DONTWAIT);
 608                 if (size != sizeof(struct worker_message))
 609                         break;
 610
 611                 /* lookup worker who sent the signal */
 612                 udev_list_node_foreach(loop, &worker_list) {
 613                         struct worker *worker = node_to_worker(loop);
 614
 615                         if (worker->pid != msg.pid)
 616                                 continue;
 617
 618                         /* worker returned */
 619                         worker->event->exitcode = msg.exitcode;
 620                         event_queue_delete(worker->event, true);
 621                         worker->event = NULL;
 622                         if (worker->state != WORKER_KILLED)
 623                                 worker->state = WORKER_IDLE;
 624                         worker_unref(worker);
 625                         break;
 626                 }
 627         }
 628 }
 629
 630 /* receive the udevd message from userspace */
 631 static struct udev_ctrl_connection *handle_ctrl_msg(struct udev_ctrl *uctrl)
 632 {
 633         struct udev *udev = udev_ctrl_get_udev(uctrl);
 634         struct udev_ctrl_connection *ctrl_conn;
 635         struct udev_ctrl_msg *ctrl_msg = NULL;
 636         const char *str;
 637         int i;
 638
 639         ctrl_conn = udev_ctrl_get_connection(uctrl);
 640         if (ctrl_conn == NULL)
 641                 goto out;
 642
 643         ctrl_msg = udev_ctrl_receive_msg(ctrl_conn);
 644         if (ctrl_msg == NULL)
 645                 goto out;
 646
 647         i = udev_ctrl_get_set_log_level(ctrl_msg);
 648         if (i >= 0) {
 649                 info(udev, "udevd message (SET_LOG_PRIORITY) received, log_priority=%i\n", i);
 650                 udev_set_log_priority(udev, i);
 651                 worker_kill(udev, 0);
 652         }
 653
 654         if (udev_ctrl_get_stop_exec_queue(ctrl_msg) > 0) {
 655                 info(udev, "udevd message (STOP_EXEC_QUEUE) received\n");
 656                 stop_exec_queue = true;
 657         }
 658
 659         if (udev_ctrl_get_start_exec_queue(ctrl_msg) > 0) {
 660                 info(udev, "udevd message (START_EXEC_QUEUE) received\n");
 661                 stop_exec_queue = false;
 662         }
 663
 664         if (udev_ctrl_get_reload_rules(ctrl_msg) > 0) {
 665                 info(udev, "udevd message (RELOAD_RULES) received\n");
 666                 reload_config = true;
 667         }
 668
 669         str = udev_ctrl_get_set_env(ctrl_msg);
 670         if (str != NULL) {
 671                 char *key;
 672
 673                 key = strdup(str);
 674                 if (key != NULL) {
 675                         char *val;
 676
 677                         val = strchr(key, '=');
 678                         if (val != NULL) {
 679                                 val[0] = '\0';
 680                                 val = &val[1];
 681                                 if (val[0] == '\0') {
 682                                         info(udev, "udevd message (ENV) received, unset '%s'\n", key);
 683                                         udev_add_property(udev, key, NULL);
 684                                 } else {
 685                                         info(udev, "udevd message (ENV) received, set '%s=%s'\n", key, val);
 686                                         udev_add_property(udev, key, val);
 687                                 }
 688                         } else {
 689                                 err(udev, "wrong key format '%s'\n", key);
 690                         }
 691                         free(key);
 692                 }
 693                 worker_kill(udev, 0);
 694         }
 695
 696         i = udev_ctrl_get_set_children_max(ctrl_msg);
 697         if (i >= 0) {
 698                 info(udev, "udevd message (SET_MAX_CHILDREN) received, children_max=%i\n", i);
 699                 children_max = i;
 700         }
 701
 702         if (udev_ctrl_get_ping(ctrl_msg) > 0)
 703                 info(udev, "udevd message (SYNC) received\n");
 704
 705         if (udev_ctrl_get_exit(ctrl_msg) > 0) {
 706                 info(udev, "udevd message (EXIT) received\n");
 707                 udev_exit = true;
 708                 /* keep reference to block the client until we exit */
 709                 udev_ctrl_connection_ref(ctrl_conn);
 710         }
 711 out:
 712         udev_ctrl_msg_unref(ctrl_msg);
 713         return udev_ctrl_connection_unref(ctrl_conn);
 714 }
 715
 716 /* read inotify messages */
 717 static int handle_inotify(struct udev *udev)
 718 {
 719         int nbytes, pos;
 720         char *buf;
 721         struct inotify_event *ev;
 722
 723         if ((ioctl(fd_inotify, FIONREAD, &nbytes) < 0) || (nbytes <= 0))
 724                 return 0;
 725
 726         buf = malloc(nbytes);
 727         if (buf == NULL) {
 728                 err(udev, "error getting buffer for inotify\n");
 729                 return -1;
 730         }
 731
 732         nbytes = read(fd_inotify, buf, nbytes);
 733
 734         for (pos = 0; pos < nbytes; pos += sizeof(struct inotify_event) + ev->len) {
 735                 struct udev_device *dev;
 736
 737                 ev = (struct inotify_event *)(buf + pos);
 738                 if (ev->len) {
 739                         const char *s;
 740
 741                         info(udev, "inotify event: %x for %s\n", ev->mask, ev->name);
 742                         s = strstr(ev->name, ".rules");
 743                         if (s == NULL)
 744                                 continue;
 745                         if (strlen(s) != strlen(".rules"))
 746                                 continue;
 747                         reload_config = true;
 748                         continue;
 749                 }
 750
 751                 dev = udev_watch_lookup(udev, ev->wd);
 752                 if (dev != NULL) {
 753                         info(udev, "inotify event: %x for %s\n", ev->mask, udev_device_get_devnode(dev));
 754                         if (ev->mask & IN_CLOSE_WRITE) {
 755                                 char filename[UTIL_PATH_SIZE];
 756                                 int fd;
 757
 758                                 info(udev, "device %s closed, synthesising 'change'\n", udev_device_get_devnode(dev));
 759                                 util_strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
 760                                 fd = open(filename, O_WRONLY);
 761                                 if (fd >= 0) {
 762                                         if (write(fd, "change", 6) < 0)
 763                                                 info(udev, "error writing uevent: %m\n");
 764                                         close(fd);
 765                                 }
 766                         }
 767                         if (ev->mask & IN_IGNORED)
 768                                 udev_watch_end(udev, dev);
 769
 770                         udev_device_unref(dev);
 771                 }
 772
 773         }
 774
 775         free(buf);
 776         return 0;
 777 }
 778
 779 static void handle_signal(struct udev *udev, int signo)
 780 {
 781         switch (signo) {
 782         case SIGINT:
 783         case SIGTERM:
 784                 udev_exit = true;
 785                 break;
 786         case SIGCHLD:
 787                 for (;;) {
 788                         pid_t pid;
 789                         int status;
 790                         struct udev_list_node *loop, *tmp;
 791
 792                         pid = waitpid(-1, &status, WNOHANG);
 793                         if (pid <= 0)
 794                                 break;
 795
 796                         udev_list_node_foreach_safe(loop, tmp, &worker_list) {
 797                                 struct worker *worker = node_to_worker(loop);
 798
 799                                 if (worker->pid != pid)
 800                                         continue;
 801                                 info(udev, "worker [%u] exit\n", pid);
 802
 803                                 if (WIFEXITED(status)) {
 804                                         if (WEXITSTATUS(status) != 0)
 805                                                 err(udev, "worker [%u] exit with return code %i\n", pid, WEXITSTATUS(status));
 806                                 } else if (WIFSIGNALED(status)) {
 807                                         err(udev, "worker [%u] terminated by signal %i (%s)\n",
 808                                             pid, WTERMSIG(status), strsignal(WTERMSIG(status)));
 809                                 } else if (WIFSTOPPED(status)) {
 810                                         err(udev, "worker [%u] stopped\n", pid);
 811                                 } else if (WIFCONTINUED(status)) {
 812                                         err(udev, "worker [%u] continued\n", pid);
 813                                 } else {
 814                                         err(udev, "worker [%u] exit with status 0x%04x\n", pid, status);
 815                                 }
 816
 817                                 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 818                                         if (worker->event != NULL) {
 819                                                 err(udev, "worker [%u] failed while handling '%s'\n",
 820                                                     pid, worker->event->devpath);
 821                                                 worker->event->exitcode = -32;
 822                                                 event_queue_delete(worker->event, true);
 823                                                 /* drop reference taken for state 'running' */
 824                                                 worker_unref(worker);
 825                                         }
 826                                 }
 827                                 worker_unref(worker);
 828                                 break;
 829                         }
 830                 }
 831                 break;
 832         case SIGHUP:
 833                 reload_config = true;
 834                 break;
 835         }
 836 }
 837
 838 static void static_dev_create_from_modules(struct udev *udev)
 839 {
 840         struct utsname kernel;
 841         char modules[UTIL_PATH_SIZE];
 842         char buf[4096];
 843         FILE *f;
 844
 845         uname(&kernel);
 846         util_strscpyl(modules, sizeof(modules), "/lib/modules/", kernel.release, "/modules.devname", NULL);
 847         f = fopen(modules, "r");
 848         if (f == NULL)
 849                 return;
 850
 851         while (fgets(buf, sizeof(buf), f) != NULL) {
 852                 char *s;
 853                 const char *modname;
 854                 const char *devname;
 855                 const char *devno;
 856                 int maj, min;
 857                 char type;
 858                 mode_t mode;
 859                 char filename[UTIL_PATH_SIZE];
 860
 861                 if (buf[0] == '#')
 862                         continue;
 863
 864                 modname = buf;
 865                 s = strchr(modname, ' ');
 866                 if (s == NULL)
 867                         continue;
 868                 s[0] = '\0';
 869
 870                 devname = &s[1];
 871                 s = strchr(devname, ' ');
 872                 if (s == NULL)
 873                         continue;
 874                 s[0] = '\0';
 875
 876                 devno = &s[1];
 877                 s = strchr(devno, ' ');
 878                 if (s == NULL)
 879                         s = strchr(devno, '\n');
 880                 if (s != NULL)
 881                         s[0] = '\0';
 882                 if (sscanf(devno, "%c%u:%u", &type, &maj, &min) != 3)
 883                         continue;
 884
 885                 /* set sticky bit, so we do not remove the node on module unload */
 886                 if (type == 'c')
 887                         mode = 01600|S_IFCHR;
 888                 else if (type == 'b')
 889                         mode = 01600|S_IFBLK;
 890                 else
 891                         continue;
 892
 893                 util_strscpyl(filename, sizeof(filename), udev_get_dev_path(udev), "/", devname, NULL);
 894                 util_create_path_selinux(udev, filename);
 895                 udev_selinux_setfscreatecon(udev, filename, mode);
 896                 info(udev, "mknod '%s' %c%u:%u\n", filename, type, maj, min);
 897                 if (mknod(filename, mode, makedev(maj, min)) < 0 && errno == EEXIST)
 898                         utimensat(AT_FDCWD, filename, NULL, 0);
 899                 udev_selinux_resetfscreatecon(udev);
 900         }
 901
 902         fclose(f);
 903 }
 904
 905 static int copy_dev_dir(struct udev *udev, DIR *dir_from, DIR *dir_to, int maxdepth)
 906 {
 907         struct dirent *dent;
 908
 909         for (dent = readdir(dir_from); dent != NULL; dent = readdir(dir_from)) {
 910                 struct stat stats;
 911
 912                 if (dent->d_name[0] == '.')
 913                         continue;
 914                 if (fstatat(dirfd(dir_from), dent->d_name, &stats, AT_SYMLINK_NOFOLLOW) != 0)
 915                         continue;
 916
 917                 if (S_ISBLK(stats.st_mode) || S_ISCHR(stats.st_mode)) {
 918                         udev_selinux_setfscreateconat(udev, dirfd(dir_to), dent->d_name, stats.st_mode & 0777);
 919                         if (mknodat(dirfd(dir_to), dent->d_name, stats.st_mode, stats.st_rdev) == 0) {
 920                                 fchmodat(dirfd(dir_to), dent->d_name, stats.st_mode & 0777, 0);
 921                                 fchownat(dirfd(dir_to), dent->d_name, stats.st_uid, stats.st_gid, 0);
 922                         } else {
 923                                 utimensat(dirfd(dir_to), dent->d_name, NULL, 0);
 924                         }
 925                         udev_selinux_resetfscreatecon(udev);
 926                 } else if (S_ISLNK(stats.st_mode)) {
 927                         char target[UTIL_PATH_SIZE];
 928                         ssize_t len;
 929
 930                         len = readlinkat(dirfd(dir_from), dent->d_name, target, sizeof(target));
 931                         if (len <= 0 || len == (ssize_t)sizeof(target))
 932                                 continue;
 933                         target[len] = '\0';
 934                         udev_selinux_setfscreateconat(udev, dirfd(dir_to), dent->d_name, S_IFLNK);
 935                         if (symlinkat(target, dirfd(dir_to), dent->d_name) < 0 && errno == EEXIST)
 936                                 utimensat(dirfd(dir_to), dent->d_name, NULL, AT_SYMLINK_NOFOLLOW);
 937                         udev_selinux_resetfscreatecon(udev);
 938                 } else if (S_ISDIR(stats.st_mode)) {
 939                         DIR *dir2_from, *dir2_to;
 940
 941                         if (maxdepth == 0)
 942                                 continue;
 943
 944                         udev_selinux_setfscreateconat(udev, dirfd(dir_to), dent->d_name, S_IFDIR|0755);
 945                         mkdirat(dirfd(dir_to), dent->d_name, 0755);
 946                         udev_selinux_resetfscreatecon(udev);
 947
 948                         dir2_to = fdopendir(openat(dirfd(dir_to), dent->d_name, O_RDONLY|O_NONBLOCK|O_DIRECTORY|O_CLOEXEC));
 949                         if (dir2_to == NULL)
 950                                 continue;
 951
 952                         dir2_from = fdopendir(openat(dirfd(dir_from), dent->d_name, O_RDONLY|O_NONBLOCK|O_DIRECTORY|O_CLOEXEC));
 953                         if (dir2_from == NULL) {
 954                                 closedir(dir2_to);
 955                                 continue;
 956                         }
 957
 958                         copy_dev_dir(udev, dir2_from, dir2_to, maxdepth-1);
 959
 960                         closedir(dir2_to);
 961                         closedir(dir2_from);
 962                 }
 963         }
 964
 965         return 0;
 966 }
 967
 968 static void static_dev_create_links(struct udev *udev, DIR *dir)
 969 {
 970         struct stdlinks {
 971                 const char *link;
 972                 const char *target;
 973         };
 974         static const struct stdlinks stdlinks[] = {
 975                 { "core", "/proc/kcore" },
 976                 { "fd", "/proc/self/fd" },
 977                 { "stdin", "/proc/self/fd/0" },
 978                 { "stdout", "/proc/self/fd/1" },
 979                 { "stderr", "/proc/self/fd/2" },
 980         };
 981         unsigned int i;
 982
 983         for (i = 0; i < ARRAY_SIZE(stdlinks); i++) {
 984                 struct stat sb;
 985
 986                 if (stat(stdlinks[i].target, &sb) == 0) {
 987                         udev_selinux_setfscreateconat(udev, dirfd(dir), stdlinks[i].link, S_IFLNK);
 988                         if (symlinkat(stdlinks[i].target, dirfd(dir), stdlinks[i].link) < 0 && errno == EEXIST)
 989                                 utimensat(dirfd(dir), stdlinks[i].link, NULL, AT_SYMLINK_NOFOLLOW);
 990                         udev_selinux_resetfscreatecon(udev);
 991                 }
 992         }
 993 }
 994
 995 static void static_dev_create_from_devices(struct udev *udev, DIR *dir)
 996 {
 997         DIR *dir_from;
 998
 999         dir_from = opendir(LIBEXECDIR "/devices");
1000         if (dir_from == NULL)
1001                 return;
1002         copy_dev_dir(udev, dir_from, dir, 8);
1003         closedir(dir_from);
1004 }
1005
1006 static void static_dev_create(struct udev *udev)
1007 {
1008         DIR *dir;
1009
1010         dir = opendir(udev_get_dev_path(udev));
1011         if (dir == NULL)
1012                 return;
1013
1014         static_dev_create_links(udev, dir);
1015         static_dev_create_from_devices(udev, dir);
1016
1017         closedir(dir);
1018 }
1019
1020 static int mem_size_mb(void)
1021 {
1022         FILE *f;
1023         char buf[4096];
1024         long int memsize = -1;
1025
1026         f = fopen("/proc/meminfo", "r");
1027         if (f == NULL)
1028                 return -1;
1029
1030         while (fgets(buf, sizeof(buf), f) != NULL) {
1031                 long int value;
1032
1033                 if (sscanf(buf, "MemTotal: %ld kB", &value) == 1) {
1034                         memsize = value / 1024;
1035                         break;
1036                 }
1037         }
1038
1039         fclose(f);
1040         return memsize;
1041 }
1042
1043 static int convert_db(struct udev *udev)
1044 {
1045         char filename[UTIL_PATH_SIZE];
1046         FILE *f;
1047         struct udev_enumerate *udev_enumerate;
1048         struct udev_list_entry *list_entry;
1049
1050         /* current database */
1051         util_strscpyl(filename, sizeof(filename), udev_get_run_path(udev), "/data", NULL);
1052         if (access(filename, F_OK) >= 0)
1053                 return 0;
1054
1055         /* make sure we do not get here again */
1056         util_create_path(udev, filename);
1057         mkdir(filename, 0755);
1058
1059         /* old database */
1060         util_strscpyl(filename, sizeof(filename), udev_get_dev_path(udev), "/.udev/db", NULL);
1061         if (access(filename, F_OK) < 0)
1062                 return 0;
1063
1064         f = fopen("/dev/kmsg", "w");
1065         if (f != NULL) {
1066                 fprintf(f, "<30>udevd[%u]: converting old udev database\n", getpid());
1067                 fclose(f);
1068         }
1069
1070         udev_enumerate = udev_enumerate_new(udev);
1071         if (udev_enumerate == NULL)
1072                 return -1;
1073         udev_enumerate_scan_devices(udev_enumerate);
1074         udev_list_entry_foreach(list_entry, udev_enumerate_get_list_entry(udev_enumerate)) {
1075                 struct udev_device *device;
1076
1077                 device = udev_device_new_from_syspath(udev, udev_list_entry_get_name(list_entry));
1078                 if (device == NULL)
1079                         continue;
1080
1081                 /* try to find the old database for devices without a current one */
1082                 if (udev_device_read_db(device, NULL) < 0) {
1083                         bool have_db;
1084                         const char *id;
1085                         struct stat stats;
1086                         char devpath[UTIL_PATH_SIZE];
1087                         char from[UTIL_PATH_SIZE];
1088
1089                         have_db = false;
1090
1091                         /* find database in old location */
1092                         id = udev_device_get_id_filename(device);
1093                         util_strscpyl(from, sizeof(from), udev_get_dev_path(udev), "/.udev/db/", id, NULL);
1094                         if (lstat(from, &stats) == 0) {
1095                                 if (!have_db) {
1096                                         udev_device_read_db(device, from);
1097                                         have_db = true;
1098                                 }
1099                                 unlink(from);
1100                         }
1101
1102                         /* find old database with $subsys:$sysname name */
1103                         util_strscpyl(from, sizeof(from), udev_get_dev_path(udev),
1104                                      "/.udev/db/", udev_device_get_subsystem(device), ":",
1105                                      udev_device_get_sysname(device), NULL);
1106                         if (lstat(from, &stats) == 0) {
1107                                 if (!have_db) {
1108                                         udev_device_read_db(device, from);
1109                                         have_db = true;
1110                                 }
1111                                 unlink(from);
1112                         }
1113
1114                         /* find old database with the encoded devpath name */
1115                         util_path_encode(udev_device_get_devpath(device), devpath, sizeof(devpath));
1116                         util_strscpyl(from, sizeof(from), udev_get_dev_path(udev), "/.udev/db/", devpath, NULL);
1117                         if (lstat(from, &stats) == 0) {
1118                                 if (!have_db) {
1119                                         udev_device_read_db(device, from);
1120                                         have_db = true;
1121                                 }
1122                                 unlink(from);
1123                         }
1124
1125                         /* write out new database */
1126                         if (have_db)
1127                                 udev_device_update_db(device);
1128                 }
1129                 udev_device_unref(device);
1130         }
1131         udev_enumerate_unref(udev_enumerate);
1132         return 0;
1133 }
1134
1135 static int systemd_fds(struct udev *udev, int *rctrl, int *rnetlink)
1136 {
1137         int ctrl = -1, netlink = -1;
1138         int fd, n;
1139
1140         n = sd_listen_fds(true);
1141         if (n <= 0)
1142                 return -1;
1143
1144         for (fd = SD_LISTEN_FDS_START; fd < n + SD_LISTEN_FDS_START; fd++) {
1145                 if (sd_is_socket(fd, AF_LOCAL, SOCK_SEQPACKET, -1)) {
1146                         if (ctrl >= 0)
1147                                 return -1;
1148                         ctrl = fd;
1149                         continue;
1150                 }
1151
1152                 if (sd_is_socket(fd, AF_NETLINK, SOCK_RAW, -1)) {
1153                         if (netlink >= 0)
1154                                 return -1;
1155                         netlink = fd;
1156                         continue;
1157                 }
1158
1159                 return -1;
1160         }
1161
1162         if (ctrl < 0 || netlink < 0)
1163                 return -1;
1164
1165         info(udev, "ctrl=%i netlink=%i\n", ctrl, netlink);
1166         *rctrl = ctrl;
1167         *rnetlink = netlink;
1168         return 0;
1169 }
1170
1171 int main(int argc, char *argv[])
1172 {
1173         struct udev *udev;
1174         FILE *f;
1175         sigset_t mask;
1176         int daemonize = false;
1177         int resolve_names = 1;
1178         static const struct option options[] = {
1179                 { "daemon", no_argument, NULL, 'd' },
1180                 { "debug", no_argument, NULL, 'D' },
1181                 { "children-max", required_argument, NULL, 'c' },
1182                 { "exec-delay", required_argument, NULL, 'e' },
1183                 { "resolve-names", required_argument, NULL, 'N' },
1184                 { "help", no_argument, NULL, 'h' },
1185                 { "version", no_argument, NULL, 'V' },
1186                 {}
1187         };
1188         int fd_ctrl = -1;
1189         int fd_netlink = -1;
1190         int fd_worker = -1;
1191         struct epoll_event ep_ctrl, ep_inotify, ep_signal, ep_netlink, ep_worker;
1192         struct udev_ctrl_connection *ctrl_conn = NULL;
1193         int rc = 1;
1194
1195         udev = udev_new();
1196         if (udev == NULL)
1197                 goto exit;
1198
1199         udev_log_init("udevd");
1200         udev_set_log_fn(udev, log_fn);
1201         info(udev, "version %s\n", VERSION);
1202         udev_selinux_init(udev);
1203
1204         /* make sure, that our runtime dir exists and is writable */
1205         if (utimensat(AT_FDCWD, udev_get_run_config_path(udev), NULL, 0) < 0) {
1206                 /* try to create our own subdirectory, do not create parent directories */
1207                 mkdir(udev_get_run_config_path(udev), 0755);
1208
1209                 if (utimensat(AT_FDCWD, udev_get_run_config_path(udev), NULL, 0) >= 0) {
1210                         /* directory seems writable now */
1211                         udev_set_run_path(udev, udev_get_run_config_path(udev));
1212                 } else {
1213                         /* fall back to /dev/.udev */
1214                         char filename[UTIL_PATH_SIZE];
1215
1216                         util_strscpyl(filename, sizeof(filename), udev_get_dev_path(udev), "/.udev", NULL);
1217                         if (udev_set_run_path(udev, filename) == NULL)
1218                                 goto exit;
1219                         mkdir(udev_get_run_path(udev), 0755);
1220                         err(udev, "error: runtime directory '%s' not writable, for now falling back to '%s'",
1221                             udev_get_run_config_path(udev), udev_get_run_path(udev));
1222                 }
1223         }
1224         /* relabel runtime dir only if it resides below /dev */
1225         if (strncmp(udev_get_run_path(udev), udev_get_dev_path(udev), strlen(udev_get_dev_path(udev))) == 0)
1226                 udev_selinux_lsetfilecon(udev, udev_get_run_path(udev), 0755);
1227         info(udev, "runtime dir '%s'\n", udev_get_run_path(udev));
1228
1229         for (;;) {
1230                 int option;
1231
1232                 option = getopt_long(argc, argv, "c:deDtN:hV", options, NULL);
1233                 if (option == -1)
1234                         break;
1235
1236                 switch (option) {
1237                 case 'd':
1238                         daemonize = true;
1239                         break;
1240                 case 'c':
1241                         children_max = strtoul(optarg, NULL, 0);
1242                         break;
1243                 case 'e':
1244                         exec_delay = strtoul(optarg, NULL, 0);
1245                         break;
1246                 case 'D':
1247                         debug = true;
1248                         if (udev_get_log_priority(udev) < LOG_INFO)
1249                                 udev_set_log_priority(udev, LOG_INFO);
1250                         break;
1251                 case 'N':
1252                         if (strcmp (optarg, "early") == 0) {
1253                                 resolve_names = 1;
1254                         } else if (strcmp (optarg, "late") == 0) {
1255                                 resolve_names = 0;
1256                         } else if (strcmp (optarg, "never") == 0) {
1257                                 resolve_names = -1;
1258                         } else {
1259                                 fprintf(stderr, "resolve-names must be early, late or never\n");
1260                                 err(udev, "resolve-names must be early, late or never\n");
1261                                 goto exit;
1262                         }
1263                         break;
1264                 case 'h':
1265                         printf("Usage: udevd OPTIONS\n"
1266                                "  --daemon\n"
1267                                "  --debug\n"
1268                                "  --children-max=<maximum number of workers>\n"
1269                                "  --exec-delay=<seconds to wait before executing RUN=>\n"
1270                                "  --resolve-names=early|late|never\n"
1271                                "  --version\n"
1272                                "  --help\n"
1273                                "\n");
1274                         goto exit;
1275                 case 'V':
1276                         printf("%s\n", VERSION);
1277                         goto exit;
1278                 default:
1279                         goto exit;
1280                 }
1281         }
1282
1283         /*
1284          * read the kernel commandline, in case we need to get into debug mode
1285          *   udev.log-priority=<level>              syslog priority
1286          *   udev.children-max=<number of workers>  events are fully serialized if set to 1
1287          *
1288          */
1289         f = fopen("/proc/cmdline", "r");
1290         if (f != NULL) {
1291                 char cmdline[4096];
1292
1293                 if (fgets(cmdline, sizeof(cmdline), f) != NULL) {
1294                         char *pos;
1295
1296                         pos = strstr(cmdline, "udev.log-priority=");
1297                         if (pos != NULL) {
1298                                 pos += strlen("udev.log-priority=");
1299                                 udev_set_log_priority(udev, util_log_priority(pos));
1300                         }
1301
1302                         pos = strstr(cmdline, "udev.children-max=");
1303                         if (pos != NULL) {
1304                                 pos += strlen("udev.children-max=");
1305                                 children_max = strtoul(pos, NULL, 0);
1306                         }
1307
1308                         pos = strstr(cmdline, "udev.exec-delay=");
1309                         if (pos != NULL) {
1310                                 pos += strlen("udev.exec-delay=");
1311                                 exec_delay = strtoul(pos, NULL, 0);
1312                         }
1313                 }
1314                 fclose(f);
1315         }
1316
1317         if (getuid() != 0) {
1318                 fprintf(stderr, "root privileges required\n");
1319                 err(udev, "root privileges required\n");
1320                 goto exit;
1321         }
1322
1323         /* set umask before creating any file/directory */
1324         chdir("/");
1325         umask(022);
1326
1327         /* create standard links, copy static nodes, create nodes from modules */
1328         static_dev_create(udev);
1329         static_dev_create_from_modules(udev);
1330
1331         /* before opening new files, make sure std{in,out,err} fds are in a sane state */
1332         if (daemonize) {
1333                 int fd;
1334
1335                 fd = open("/dev/null", O_RDWR);
1336                 if (fd >= 0) {
1337                         if (write(STDOUT_FILENO, 0, 0) < 0)
1338                                 dup2(fd, STDOUT_FILENO);
1339                         if (write(STDERR_FILENO, 0, 0) < 0)
1340                                 dup2(fd, STDERR_FILENO);
1341                         if (fd > STDERR_FILENO)
1342                                 close(fd);
1343                 } else {
1344                         fprintf(stderr, "cannot open /dev/null\n");
1345                         err(udev, "cannot open /dev/null\n");
1346                 }
1347         }
1348
1349         if (systemd_fds(udev, &fd_ctrl, &fd_netlink) >= 0) {
1350                 /* get control and netlink socket from from systemd */
1351                 udev_ctrl = udev_ctrl_new_from_socket_fd(udev, UDEV_CTRL_SOCK_PATH, fd_ctrl);
1352                 if (udev_ctrl == NULL) {
1353                         err(udev, "error taking over udev control socket");
1354                         rc = 1;
1355                         goto exit;
1356                 }
1357
1358                 monitor = udev_monitor_new_from_netlink_fd(udev, "kernel", fd_netlink);
1359                 if (monitor == NULL) {
1360                         err(udev, "error taking over netlink socket\n");
1361                         rc = 3;
1362                         goto exit;
1363                 }
1364         } else {
1365                 /* open control and netlink socket */
1366                 udev_ctrl = udev_ctrl_new_from_socket(udev, UDEV_CTRL_SOCK_PATH);
1367                 if (udev_ctrl == NULL) {
1368                         fprintf(stderr, "error initializing udev control socket");
1369                         err(udev, "error initializing udev control socket");
1370                         rc = 1;
1371                         goto exit;
1372                 }
1373                 fd_ctrl = udev_ctrl_get_fd(udev_ctrl);
1374
1375                 monitor = udev_monitor_new_from_netlink(udev, "kernel");
1376                 if (monitor == NULL) {
1377                         fprintf(stderr, "error initializing netlink socket\n");
1378                         err(udev, "error initializing netlink socket\n");
1379                         rc = 3;
1380                         goto exit;
1381                 }
1382                 fd_netlink = udev_monitor_get_fd(monitor);
1383         }
1384
1385         if (udev_monitor_enable_receiving(monitor) < 0) {
1386                 fprintf(stderr, "error binding netlink socket\n");
1387                 err(udev, "error binding netlink socket\n");
1388                 rc = 3;
1389                 goto exit;
1390         }
1391
1392         if (udev_ctrl_enable_receiving(udev_ctrl) < 0) {
1393                 fprintf(stderr, "error binding udev control socket\n");
1394                 err(udev, "error binding udev control socket\n");
1395                 rc = 1;
1396                 goto exit;
1397         }
1398
1399         udev_monitor_set_receive_buffer_size(monitor, 128*1024*1024);
1400
1401         /* create queue file before signalling 'ready', to make sure we block 'settle' */
1402         udev_queue_export = udev_queue_export_new(udev);
1403         if (udev_queue_export == NULL) {
1404                 err(udev, "error creating queue file\n");
1405                 goto exit;
1406         }
1407
1408         if (daemonize) {
1409                 pid_t pid;
1410                 int fd;
1411
1412                 pid = fork();
1413                 switch (pid) {
1414                 case 0:
1415                         break;
1416                 case -1:
1417                         err(udev, "fork of daemon failed: %m\n");
1418                         rc = 4;
1419                         goto exit;
1420                 default:
1421                         rc = EXIT_SUCCESS;
1422                         goto exit_keep_queue;
1423                 }
1424
1425                 setsid();
1426
1427                 fd = open("/proc/self/oom_score_adj", O_RDWR);
1428                 if (fd < 0) {
1429                         /* Fallback to old interface */
1430                         fd = open("/proc/self/oom_adj", O_RDWR);
1431                         if (fd < 0) {
1432                                 err(udev, "error disabling OOM: %m\n");
1433                         } else {
1434                                 /* OOM_DISABLE == -17 */
1435                                 write(fd, "-17", 3);
1436                                 close(fd);
1437                         }
1438                 } else {
1439                         write(fd, "-1000", 5);
1440                         close(fd);
1441                 }
1442         } else {
1443                 sd_notify(1, "READY=1");
1444         }
1445
1446         f = fopen("/dev/kmsg", "w");
1447         if (f != NULL) {
1448                 fprintf(f, "<30>udevd[%u]: starting version " VERSION "\n", getpid());
1449                 fclose(f);
1450         }
1451
1452         if (!debug) {
1453                 int fd;
1454
1455                 fd = open("/dev/null", O_RDWR);
1456                 if (fd >= 0) {
1457                         dup2(fd, STDIN_FILENO);
1458                         dup2(fd, STDOUT_FILENO);
1459                         dup2(fd, STDERR_FILENO);
1460                         close(fd);
1461                 }
1462         }
1463
1464         fd_inotify = udev_watch_init(udev);
1465         if (fd_inotify < 0) {
1466                 fprintf(stderr, "error initializing inotify\n");
1467                 err(udev, "error initializing inotify\n");
1468                 rc = 4;
1469                 goto exit;
1470         }
1471
1472         if (udev_get_rules_path(udev) != NULL) {
1473                 inotify_add_watch(fd_inotify, udev_get_rules_path(udev),
1474                                   IN_DELETE | IN_MOVE | IN_CLOSE_WRITE);
1475         } else {
1476                 char filename[UTIL_PATH_SIZE];
1477                 struct stat statbuf;
1478
1479                 inotify_add_watch(fd_inotify, LIBEXECDIR "/rules.d",
1480                                   IN_DELETE | IN_MOVE | IN_CLOSE_WRITE);
1481                 inotify_add_watch(fd_inotify, SYSCONFDIR "/udev/rules.d",
1482                                   IN_DELETE | IN_MOVE | IN_CLOSE_WRITE);
1483
1484                 /* watch dynamic rules directory */
1485                 util_strscpyl(filename, sizeof(filename), udev_get_run_path(udev), "/rules.d", NULL);
1486                 if (stat(filename, &statbuf) != 0) {
1487                         util_create_path(udev, filename);
1488                         mkdir(filename, 0755);
1489                 }
1490                 inotify_add_watch(fd_inotify, filename,
1491                                   IN_DELETE | IN_MOVE | IN_CLOSE_WRITE);
1492         }
1493         udev_watch_restore(udev);
1494
1495         /* block and listen to all signals on signalfd */
1496         sigfillset(&mask);
1497         sigprocmask(SIG_SETMASK, &mask, &sigmask_orig);
1498         fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
1499         if (fd_signal < 0) {
1500                 fprintf(stderr, "error creating signalfd\n");
1501                 err(udev, "error creating signalfd\n");
1502                 rc = 5;
1503                 goto exit;
1504         }
1505
1506         /* unnamed socket from workers to the main daemon */
1507         if (socketpair(AF_LOCAL, SOCK_DGRAM|SOCK_CLOEXEC, 0, worker_watch) < 0) {
1508                 fprintf(stderr, "error creating socketpair\n");
1509                 err(udev, "error creating socketpair\n");
1510                 rc = 6;
1511                 goto exit;
1512         }
1513         fd_worker = worker_watch[READ_END];
1514
1515         rules = udev_rules_new(udev, resolve_names);
1516         if (rules == NULL) {
1517                 err(udev, "error reading rules\n");
1518                 goto exit;
1519         }
1520
1521         memset(&ep_ctrl, 0, sizeof(struct epoll_event));
1522         ep_ctrl.events = EPOLLIN;
1523         ep_ctrl.data.fd = fd_ctrl;
1524
1525         memset(&ep_inotify, 0, sizeof(struct epoll_event));
1526         ep_inotify.events = EPOLLIN;
1527         ep_inotify.data.fd = fd_inotify;
1528
1529         memset(&ep_signal, 0, sizeof(struct epoll_event));
1530         ep_signal.events = EPOLLIN;
1531         ep_signal.data.fd = fd_signal;
1532
1533         memset(&ep_netlink, 0, sizeof(struct epoll_event));
1534         ep_netlink.events = EPOLLIN;
1535         ep_netlink.data.fd = fd_netlink;
1536
1537         memset(&ep_worker, 0, sizeof(struct epoll_event));
1538         ep_worker.events = EPOLLIN;
1539         ep_worker.data.fd = fd_worker;
1540
1541         fd_ep = epoll_create1(EPOLL_CLOEXEC);
1542         if (fd_ep < 0) {
1543                 err(udev, "error creating epoll fd: %m\n");
1544                 goto exit;
1545         }
1546         if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_ctrl, &ep_ctrl) < 0 ||
1547             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_inotify, &ep_inotify) < 0 ||
1548             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
1549             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_netlink, &ep_netlink) < 0 ||
1550             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_worker, &ep_worker) < 0) {
1551                 err(udev, "fail to add fds to epoll: %m\n");
1552                 goto exit;
1553         }
1554
1555         /* if needed, convert old database from earlier udev version */
1556         convert_db(udev);
1557
1558         if (children_max <= 0) {
1559                 int memsize = mem_size_mb();
1560
1561                 /* set value depending on the amount of RAM */
1562                 if (memsize > 0)
1563                         children_max = 128 + (memsize / 8);
1564                 else
1565                         children_max = 128;
1566         }
1567         info(udev, "set children_max to %u\n", children_max);
1568
1569         udev_rules_apply_static_dev_perms(rules);
1570
1571         udev_list_node_init(&event_list);
1572         udev_list_node_init(&worker_list);
1573
1574         for (;;) {
1575                 struct epoll_event ev[8];
1576                 int fdcount;
1577                 int timeout;
1578                 bool is_worker, is_signal, is_inotify, is_netlink, is_ctrl;
1579                 int i;
1580
1581                 if (udev_exit) {
1582                         /* close sources of new events and discard buffered events */
1583                         if (fd_ctrl >= 0) {
1584                                 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_ctrl, NULL);
1585                                 fd_ctrl = -1;
1586                         }
1587                         if (monitor != NULL) {
1588                                 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_netlink, NULL);
1589                                 udev_monitor_unref(monitor);
1590                                 monitor = NULL;
1591                         }
1592                         if (fd_inotify >= 0) {
1593                                 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_inotify, NULL);
1594                                 close(fd_inotify);
1595                                 fd_inotify = -1;
1596                         }
1597
1598                         /* discard queued events and kill workers */
1599                         event_queue_cleanup(udev, EVENT_QUEUED);
1600                         worker_kill(udev, 0);
1601
1602                         /* exit after all has cleaned up */
1603                         if (udev_list_node_is_empty(&event_list) && udev_list_node_is_empty(&worker_list))
1604                                 break;
1605
1606                         /* timeout at exit for workers to finish */
1607                         timeout = 60 * 1000;
1608                 } else if (udev_list_node_is_empty(&event_list) && children > 2) {
1609                         /* set timeout to kill idle workers */
1610                         timeout = 3 * 1000;
1611                 } else {
1612                         timeout = -1;
1613                 }
1614                 fdcount = epoll_wait(fd_ep, ev, ARRAY_SIZE(ev), timeout);
1615                 if (fdcount < 0)
1616                         continue;
1617
1618                 if (fdcount == 0) {
1619                         if (udev_exit) {
1620                                 info(udev, "timeout, giving up waiting for workers to finish\n");
1621                                 break;
1622                         }
1623
1624                         /* timeout - kill idle workers */
1625                         worker_kill(udev, 2);
1626                 }
1627
1628                 is_worker = is_signal = is_inotify = is_netlink = is_ctrl = false;
1629                 for (i = 0; i < fdcount; i++) {
1630                         if (ev[i].data.fd == fd_worker && ev[i].events & EPOLLIN)
1631                                 is_worker = true;
1632                         else if (ev[i].data.fd == fd_netlink && ev[i].events & EPOLLIN)
1633                                 is_netlink = true;
1634                         else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN)
1635                                 is_signal = true;
1636                         else if (ev[i].data.fd == fd_inotify && ev[i].events & EPOLLIN)
1637                                 is_inotify = true;
1638                         else if (ev[i].data.fd == fd_ctrl && ev[i].events & EPOLLIN)
1639                                 is_ctrl = true;
1640                 }
1641
1642                 /* event has finished */
1643                 if (is_worker)
1644                         worker_returned(fd_worker);
1645
1646                 if (is_netlink) {
1647                         struct udev_device *dev;
1648
1649                         dev = udev_monitor_receive_device(monitor);
1650                         if (dev != NULL)
1651                                 udev_device_set_usec_initialized(dev, now_usec());
1652                                 if (event_queue_insert(dev) < 0)
1653                                         udev_device_unref(dev);
1654                 }
1655
1656                 /* start new events */
1657                 if (!udev_list_node_is_empty(&event_list) && !udev_exit && !stop_exec_queue)
1658                         event_queue_start(udev);
1659
1660                 if (is_signal) {
1661                         struct signalfd_siginfo fdsi;
1662                         ssize_t size;
1663
1664                         size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
1665                         if (size == sizeof(struct signalfd_siginfo))
1666                                 handle_signal(udev, fdsi.ssi_signo);
1667                 }
1668
1669                 /* we are shutting down, the events below are not handled anymore */
1670                 if (udev_exit)
1671                         continue;
1672
1673                 /* device node and rules directory inotify watch */
1674                 if (is_inotify)
1675                         handle_inotify(udev);
1676
1677                 /*
1678                  * This needs to be after the inotify handling, to make sure,
1679                  * that the ping is send back after the possibly generated
1680                  * "change" events by the inotify device node watch.
1681                  *
1682                  * A single time we may receive a client connection which we need to
1683                  * keep open to block the client. It will be closed right before we
1684                  * exit.
1685                  */
1686                 if (is_ctrl)
1687                         ctrl_conn = handle_ctrl_msg(udev_ctrl);
1688
1689                 /* rules changed, set by inotify or a HUP signal */
1690                 if (reload_config) {
1691                         struct udev_rules *rules_new;
1692
1693                         worker_kill(udev, 0);
1694                         rules_new = udev_rules_new(udev, resolve_names);
1695                         if (rules_new != NULL) {
1696                                 udev_rules_unref(rules);
1697                                 rules = rules_new;
1698                         }
1699                         reload_config = 0;
1700                 }
1701         }
1702
1703         rc = EXIT_SUCCESS;
1704 exit:
1705         udev_queue_export_cleanup(udev_queue_export);
1706 exit_keep_queue:
1707         if (fd_ep >= 0)
1708                 close(fd_ep);
1709         worker_list_cleanup(udev);
1710         event_queue_cleanup(udev, EVENT_UNDEF);
1711         udev_rules_unref(rules);
1712         if (fd_signal >= 0)
1713                 close(fd_signal);
1714         if (worker_watch[READ_END] >= 0)
1715                 close(worker_watch[READ_END]);
1716         if (worker_watch[WRITE_END] >= 0)
1717                 close(worker_watch[WRITE_END]);
1718         udev_monitor_unref(monitor);
1719         udev_queue_export_unref(udev_queue_export);
1720         udev_ctrl_connection_unref(ctrl_conn);
1721         udev_ctrl_unref(udev_ctrl);
1722         udev_selinux_exit(udev);
1723         udev_unref(udev);
1724         udev_log_close();
1725         return rc;
1726 }