udev/udevd.c

   1 /*
   2  * Copyright (C) 2004-2009 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/signalfd.h>
  38 #include <sys/select.h>
  39 #include <sys/poll.h>
  40 #include <sys/wait.h>
  41 #include <sys/stat.h>
  42 #include <sys/ioctl.h>
  43 #include <sys/inotify.h>
  44
  45 #include "udev.h"
  46
  47 #define UDEVD_PRIORITY                  -4
  48 #define UDEV_PRIORITY                   -2
  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 timeval tv;
  59                 struct timezone tz;
  60
  61                 vsnprintf(buf, sizeof(buf), format, args);
  62                 gettimeofday(&tv, &tz);
  63                 fprintf(stderr, "%llu.%06u [%u] %s: %s",
  64                         (unsigned long long) tv.tv_sec, (unsigned int) tv.tv_usec,
  65                         (int) getpid(), fn, buf);
  66         } else {
  67                 vsyslog(priority, format, args);
  68         }
  69 }
  70
  71 static bool debug_trace;
  72 static struct udev_rules *rules;
  73 static struct udev_queue_export *udev_queue_export;
  74 static struct udev_ctrl *udev_ctrl;
  75 static struct udev_monitor *monitor;
  76 static int worker_watch[2];
  77 static pid_t settle_pid;
  78 static bool stop_exec_queue;
  79 static bool reload_config;
  80 static int max_childs;
  81 static int childs;
  82 static sigset_t orig_sigmask;
  83 static struct udev_list_node event_list;
  84 static struct udev_list_node worker_list;
  85 static bool udev_exit;
  86 static volatile sig_atomic_t worker_exit;
  87
  88 enum poll_fd {
  89         FD_CONTROL,
  90         FD_NETLINK,
  91         FD_INOTIFY,
  92         FD_SIGNAL,
  93         FD_WORKER,
  94 };
  95
  96 static struct pollfd pfd[] = {
  97         [FD_NETLINK] = { .events = POLLIN },
  98         [FD_WORKER] =  { .events = POLLIN },
  99         [FD_SIGNAL] =  { .events = POLLIN },
 100         [FD_INOTIFY] = { .events = POLLIN },
 101         [FD_CONTROL] = { .events = POLLIN },
 102 };
 103
 104 enum event_state {
 105         EVENT_UNDEF,
 106         EVENT_QUEUED,
 107         EVENT_RUNNING,
 108 };
 109
 110 struct event {
 111         struct udev_list_node node;
 112         struct udev *udev;
 113         struct udev_device *dev;
 114         enum event_state state;
 115         int exitcode;
 116         unsigned long long int delaying_seqnum;
 117         unsigned long long int seqnum;
 118         const char *devpath;
 119         size_t devpath_len;
 120         const char *devpath_old;
 121         dev_t devnum;
 122         bool is_block;
 123 };
 124
 125 static struct event *node_to_event(struct udev_list_node *node)
 126 {
 127         char *event;
 128
 129         event = (char *)node;
 130         event -= offsetof(struct event, node);
 131         return (struct event *)event;
 132 }
 133
 134 enum worker_state {
 135         WORKER_UNDEF,
 136         WORKER_RUNNING,
 137         WORKER_IDLE,
 138         WORKER_KILLED,
 139 };
 140
 141 struct worker {
 142         struct udev_list_node node;
 143         struct udev *udev;
 144         int refcount;
 145         pid_t pid;
 146         struct udev_monitor *monitor;
 147         enum worker_state state;
 148         struct event *event;
 149 };
 150
 151 /* passed from worker to main process */
 152 struct worker_message {
 153         pid_t pid;
 154         int exitcode;
 155 };
 156
 157 static struct worker *node_to_worker(struct udev_list_node *node)
 158 {
 159         char *worker;
 160
 161         worker = (char *)node;
 162         worker -= offsetof(struct worker, node);
 163         return (struct worker *)worker;
 164 }
 165
 166 static void event_queue_delete(struct event *event)
 167 {
 168         udev_list_node_remove(&event->node);
 169
 170         /* mark as failed, if "add" event returns non-zero */
 171         if (event->exitcode != 0 && strcmp(udev_device_get_action(event->dev), "remove") != 0)
 172                 udev_queue_export_device_failed(udev_queue_export, event->dev);
 173         else
 174                 udev_queue_export_device_finished(udev_queue_export, event->dev);
 175
 176         info(event->udev, "seq %llu done with %i\n", udev_device_get_seqnum(event->dev), event->exitcode);
 177         udev_device_unref(event->dev);
 178         free(event);
 179 }
 180
 181 static void event_sig_handler(int signum)
 182 {
 183         switch (signum) {
 184         case SIGALRM:
 185                 _exit(1);
 186                 break;
 187         case SIGTERM:
 188                 worker_exit = true;
 189                 break;
 190         }
 191 }
 192
 193 static struct worker *worker_ref(struct worker *worker)
 194 {
 195         worker->refcount++;
 196         return worker;
 197 }
 198
 199 static void worker_unref(struct worker *worker)
 200 {
 201         worker->refcount--;
 202         if (worker->refcount > 0)
 203                 return;
 204
 205         udev_list_node_remove(&worker->node);
 206         udev_monitor_unref(worker->monitor);
 207         childs--;
 208         info(worker->udev, "worker [%u] cleaned up\n", worker->pid);
 209         free(worker);
 210 }
 211
 212 static void worker_new(struct event *event)
 213 {
 214         struct worker *worker;
 215         struct udev_monitor *worker_monitor;
 216         pid_t pid;
 217         struct sigaction act;
 218
 219         /* listen for new events */
 220         worker_monitor = udev_monitor_new_from_netlink(event->udev, NULL);
 221         if (worker_monitor == NULL)
 222                 return;
 223         /* allow the main daemon netlink address to send devices to the worker */
 224         udev_monitor_allow_unicast_sender(worker_monitor, monitor);
 225         udev_monitor_enable_receiving(worker_monitor);
 226
 227         worker = calloc(1, sizeof(struct worker));
 228         if (worker == NULL)
 229                 return;
 230         /* worker + event reference */
 231         worker->refcount = 2;
 232         worker->udev = event->udev;
 233
 234         pid = fork();
 235         switch (pid) {
 236         case 0: {
 237                 sigset_t sigmask;
 238                 struct udev_device *dev;
 239                 struct pollfd pmon = {
 240                         .fd = udev_monitor_get_fd(worker_monitor),
 241                         .events = POLLIN,
 242                 };
 243
 244                 udev_queue_export_unref(udev_queue_export);
 245                 udev_monitor_unref(monitor);
 246                 udev_ctrl_unref(udev_ctrl);
 247                 close(pfd[FD_SIGNAL].fd);
 248                 close(worker_watch[READ_END]);
 249                 udev_log_close();
 250                 udev_log_init("udevd-work");
 251                 setpriority(PRIO_PROCESS, 0, UDEV_PRIORITY);
 252
 253                 /* set signal handlers */
 254                 memset(&act, 0x00, sizeof(act));
 255                 act.sa_handler = event_sig_handler;
 256                 sigemptyset (&act.sa_mask);
 257                 act.sa_flags = 0;
 258                 sigaction(SIGTERM, &act, NULL);
 259                 sigaction(SIGALRM, &act, NULL);
 260
 261                 /* unblock SIGALRM */
 262                 sigfillset(&sigmask);
 263                 sigdelset(&sigmask, SIGALRM);
 264                 sigprocmask(SIG_SETMASK, &sigmask, NULL);
 265                 /* SIGTERM is unblocked in ppoll() */
 266                 sigdelset(&sigmask, SIGTERM);
 267
 268                 /* request TERM signal if parent exits */
 269                 prctl(PR_SET_PDEATHSIG, SIGTERM);
 270
 271                 /* initial device */
 272                 dev = event->dev;
 273
 274                 do {
 275                         struct udev_event *udev_event;
 276                         struct worker_message msg = {};
 277                         int err;
 278                         int failed = 0;
 279
 280                         info(event->udev, "seq %llu running\n", udev_device_get_seqnum(dev));
 281                         udev_event = udev_event_new(dev);
 282                         if (udev_event == NULL)
 283                                 _exit(3);
 284
 285                         /* set timeout to prevent hanging processes */
 286                         alarm(UDEV_EVENT_TIMEOUT);
 287
 288                         /* apply rules, create node, symlinks */
 289                         err = udev_event_execute_rules(udev_event, rules);
 290
 291                         /* rules may change/disable the timeout */
 292                         if (udev_device_get_event_timeout(dev) >= 0)
 293                                 alarm(udev_device_get_event_timeout(dev));
 294
 295                         if (err == 0)
 296                                 failed = udev_event_execute_run(udev_event, &orig_sigmask);
 297
 298                         alarm(0);
 299
 300                         /* apply/restore inotify watch */
 301                         if (err == 0 && udev_event->inotify_watch) {
 302                                 udev_watch_begin(udev_event->udev, dev);
 303                                 udev_device_update_db(dev);
 304                         }
 305
 306                         /* send processed event back to libudev listeners */
 307                         udev_monitor_send_device(worker_monitor, NULL, dev);
 308
 309                         /* send udevd the result of the event execution */
 310                         if (err != 0)
 311                                 msg.exitcode = err;
 312                         else if (failed != 0)
 313                                 msg.exitcode = failed;
 314                         msg.pid = getpid();
 315                         send(worker_watch[WRITE_END], &msg, sizeof(struct worker_message), 0);
 316
 317                         info(event->udev, "seq %llu processed with %i\n", udev_device_get_seqnum(dev), err);
 318                         udev_event_unref(udev_event);
 319                         udev_device_unref(dev);
 320                         dev = NULL;
 321
 322                         /* wait for more device messages or signal from udevd */
 323                         while (!worker_exit) {
 324                                 int fdcount;
 325
 326                                 fdcount = ppoll(&pmon, 1, NULL, &sigmask);
 327                                 if (fdcount < 0)
 328                                         continue;
 329
 330                                 if (pmon.revents & POLLIN) {
 331                                         dev = udev_monitor_receive_device(worker_monitor);
 332                                         if (dev != NULL)
 333                                                 break;
 334                                 }
 335                         }
 336                 } while (dev != NULL);
 337
 338                 udev_monitor_unref(worker_monitor);
 339                 udev_log_close();
 340                 exit(0);
 341         }
 342         case -1:
 343                 udev_monitor_unref(worker_monitor);
 344                 event->state = EVENT_QUEUED;
 345                 free(worker);
 346                 err(event->udev, "fork of child failed: %m\n");
 347                 break;
 348         default:
 349                 /* close monitor, but keep address around */
 350                 udev_monitor_disconnect(worker_monitor);
 351                 worker->monitor = worker_monitor;
 352                 worker->pid = pid;
 353                 worker->state = WORKER_RUNNING;
 354                 worker->event = event;
 355                 event->state = EVENT_RUNNING;
 356                 udev_list_node_append(&worker->node, &worker_list);
 357                 childs++;
 358                 info(event->udev, "seq %llu forked new worker [%u]\n", udev_device_get_seqnum(event->dev), pid);
 359                 break;
 360         }
 361 }
 362
 363 static void event_run(struct event *event, bool force)
 364 {
 365         struct udev_list_node *loop;
 366
 367         udev_list_node_foreach(loop, &worker_list) {
 368                 struct worker *worker = node_to_worker(loop);
 369                 ssize_t count;
 370
 371                 if (worker->state != WORKER_IDLE)
 372                         continue;
 373
 374                 count = udev_monitor_send_device(monitor, worker->monitor, event->dev);
 375                 if (count < 0) {
 376                         err(event->udev, "worker [%u] did not accept message %zi (%m), kill it\n", worker->pid, count);
 377                         kill(worker->pid, SIGKILL);
 378                         worker->state = WORKER_KILLED;
 379                         continue;
 380                 }
 381                 worker_ref(worker);
 382                 worker->event = event;
 383                 worker->state = WORKER_RUNNING;
 384                 event->state = EVENT_RUNNING;
 385                 return;
 386         }
 387
 388         if (!force && childs >= max_childs) {
 389                 info(event->udev, "maximum number (%i) of childs reached\n", childs);
 390                 return;
 391         }
 392
 393         /* start new worker and pass initial device */
 394         worker_new(event);
 395 }
 396
 397 static void event_queue_insert(struct udev_device *dev)
 398 {
 399         struct event *event;
 400
 401         event = calloc(1, sizeof(struct event));
 402         if (event == NULL)
 403                 return;
 404
 405         event->udev = udev_device_get_udev(dev);
 406         event->dev = dev;
 407         event->seqnum = udev_device_get_seqnum(dev);
 408         event->devpath = udev_device_get_devpath(dev);
 409         event->devpath_len = strlen(event->devpath);
 410         event->devpath_old = udev_device_get_devpath_old(dev);
 411         event->devnum = udev_device_get_devnum(dev);
 412         event->is_block = (strcmp("block", udev_device_get_subsystem(dev)) == 0);
 413
 414         udev_queue_export_device_queued(udev_queue_export, dev);
 415         info(event->udev, "seq %llu queued, '%s' '%s'\n", udev_device_get_seqnum(dev),
 416              udev_device_get_action(dev), udev_device_get_subsystem(dev));
 417
 418         event->state = EVENT_QUEUED;
 419         udev_list_node_append(&event->node, &event_list);
 420
 421         /* run all events with a timeout set immediately */
 422         if (udev_device_get_timeout(dev) > 0) {
 423                 event_run(event, true);
 424                 return;
 425         }
 426 }
 427
 428 static void worker_kill(struct udev *udev, int retain)
 429 {
 430         struct udev_list_node *loop;
 431         int max;
 432
 433         if (childs <= retain)
 434                 return;
 435
 436         max = childs - retain;
 437
 438         udev_list_node_foreach(loop, &worker_list) {
 439                 struct worker *worker = node_to_worker(loop);
 440
 441                 if (max-- <= 0)
 442                         break;
 443
 444                 if (worker->state == WORKER_KILLED)
 445                         continue;
 446
 447                 worker->state = WORKER_KILLED;
 448                 kill(worker->pid, SIGTERM);
 449         }
 450 }
 451
 452 /* lookup event for identical, parent, child device */
 453 static bool is_devpath_busy(struct event *event)
 454 {
 455         struct udev_list_node *loop;
 456         size_t common;
 457
 458         /* check if queue contains events we depend on */
 459         udev_list_node_foreach(loop, &event_list) {
 460                 struct event *loop_event = node_to_event(loop);
 461
 462                 /* we already found a later event, earlier can not block us, no need to check again */
 463                 if (loop_event->seqnum < event->delaying_seqnum)
 464                         continue;
 465
 466                 /* event we checked earlier still exists, no need to check again */
 467                 if (loop_event->seqnum == event->delaying_seqnum)
 468                         return true;
 469
 470                 /* found ourself, no later event can block us */
 471                 if (loop_event->seqnum >= event->seqnum)
 472                         break;
 473
 474                 /* check major/minor */
 475                 if (major(event->devnum) != 0 && event->devnum == loop_event->devnum && event->is_block == loop_event->is_block)
 476                         return true;
 477
 478                 /* check our old name */
 479                 if (event->devpath_old != NULL && strcmp(loop_event->devpath, event->devpath_old) == 0) {
 480                         event->delaying_seqnum = loop_event->seqnum;
 481                         return true;
 482                 }
 483
 484                 /* compare devpath */
 485                 common = MIN(loop_event->devpath_len, event->devpath_len);
 486
 487                 /* one devpath is contained in the other? */
 488                 if (memcmp(loop_event->devpath, event->devpath, common) != 0)
 489                         continue;
 490
 491                 /* identical device event found */
 492                 if (loop_event->devpath_len == event->devpath_len) {
 493                         event->delaying_seqnum = loop_event->seqnum;
 494                         return true;
 495                 }
 496
 497                 /* parent device event found */
 498                 if (event->devpath[common] == '/') {
 499                         event->delaying_seqnum = loop_event->seqnum;
 500                         return true;
 501                 }
 502
 503                 /* child device event found */
 504                 if (loop_event->devpath[common] == '/') {
 505                         event->delaying_seqnum = loop_event->seqnum;
 506                         return true;
 507                 }
 508
 509                 /* no matching device */
 510                 continue;
 511         }
 512
 513         return false;
 514 }
 515
 516 static void events_start(struct udev *udev)
 517 {
 518         struct udev_list_node *loop;
 519
 520         udev_list_node_foreach(loop, &event_list) {
 521                 struct event *event = node_to_event(loop);
 522
 523                 if (event->state != EVENT_QUEUED)
 524                         continue;
 525
 526                 /* do not start event if parent or child event is still running */
 527                 if (is_devpath_busy(event)) {
 528                         dbg(udev, "delay seq %llu (%s)\n", event->seqnum, event->devpath);
 529                         continue;
 530                 }
 531
 532                 event_run(event, false);
 533         }
 534 }
 535
 536 static void worker_returned(void)
 537 {
 538         while (1) {
 539                 struct worker_message msg;
 540                 ssize_t size;
 541                 struct udev_list_node *loop;
 542
 543                 size = recv(pfd[FD_WORKER].fd, &msg, sizeof(struct worker_message), MSG_DONTWAIT);
 544                 if (size != sizeof(struct worker_message))
 545                         break;
 546
 547                 /* lookup worker who sent the signal */
 548                 udev_list_node_foreach(loop, &worker_list) {
 549                         struct worker *worker = node_to_worker(loop);
 550
 551                         if (worker->pid != msg.pid)
 552                                 continue;
 553
 554                         /* worker returned */
 555                         worker->event->exitcode = msg.exitcode;
 556                         event_queue_delete(worker->event);
 557                         worker->event = NULL;
 558                         if (worker->state != WORKER_KILLED)
 559                                 worker->state = WORKER_IDLE;
 560                         worker_unref(worker);
 561                         break;
 562                 }
 563         }
 564 }
 565
 566 /* receive the udevd message from userspace */
 567 static void handle_ctrl_msg(struct udev_ctrl *uctrl)
 568 {
 569         struct udev *udev = udev_ctrl_get_udev(uctrl);
 570         struct udev_ctrl_msg *ctrl_msg;
 571         const char *str;
 572         int i;
 573
 574         ctrl_msg = udev_ctrl_receive_msg(uctrl);
 575         if (ctrl_msg == NULL)
 576                 return;
 577
 578         i = udev_ctrl_get_set_log_level(ctrl_msg);
 579         if (i >= 0) {
 580                 info(udev, "udevd message (SET_LOG_PRIORITY) received, log_priority=%i\n", i);
 581                 udev_set_log_priority(udev, i);
 582                 worker_kill(udev, 0);
 583         }
 584
 585         if (udev_ctrl_get_stop_exec_queue(ctrl_msg) > 0) {
 586                 info(udev, "udevd message (STOP_EXEC_QUEUE) received\n");
 587                 stop_exec_queue = true;
 588         }
 589
 590         if (udev_ctrl_get_start_exec_queue(ctrl_msg) > 0) {
 591                 info(udev, "udevd message (START_EXEC_QUEUE) received\n");
 592                 stop_exec_queue = false;
 593         }
 594
 595         if (udev_ctrl_get_reload_rules(ctrl_msg) > 0) {
 596                 info(udev, "udevd message (RELOAD_RULES) received\n");
 597                 reload_config = true;
 598         }
 599
 600         str = udev_ctrl_get_set_env(ctrl_msg);
 601         if (str != NULL) {
 602                 char *key;
 603
 604                 key = strdup(str);
 605                 if (key != NULL) {
 606                         char *val;
 607
 608                         val = strchr(key, '=');
 609                         if (val != NULL) {
 610                                 val[0] = '\0';
 611                                 val = &val[1];
 612                                 if (val[0] == '\0') {
 613                                         info(udev, "udevd message (ENV) received, unset '%s'\n", key);
 614                                         udev_add_property(udev, key, NULL);
 615                                 } else {
 616                                         info(udev, "udevd message (ENV) received, set '%s=%s'\n", key, val);
 617                                         udev_add_property(udev, key, val);
 618                                 }
 619                         } else {
 620                                 err(udev, "wrong key format '%s'\n", key);
 621                         }
 622                         free(key);
 623                 }
 624                 worker_kill(udev, 0);
 625         }
 626
 627         i = udev_ctrl_get_set_max_childs(ctrl_msg);
 628         if (i >= 0) {
 629                 info(udev, "udevd message (SET_MAX_CHILDS) received, max_childs=%i\n", i);
 630                 max_childs = i;
 631         }
 632
 633         settle_pid = udev_ctrl_get_settle(ctrl_msg);
 634         if (settle_pid > 0) {
 635                 info(udev, "udevd message (SETTLE) received\n");
 636                 kill(settle_pid, SIGUSR1);
 637                 settle_pid = 0;
 638         }
 639         udev_ctrl_msg_unref(ctrl_msg);
 640 }
 641
 642 /* read inotify messages */
 643 static int handle_inotify(struct udev *udev)
 644 {
 645         int nbytes, pos;
 646         char *buf;
 647         struct inotify_event *ev;
 648
 649         if ((ioctl(pfd[FD_INOTIFY].fd, FIONREAD, &nbytes) < 0) || (nbytes <= 0))
 650                 return 0;
 651
 652         buf = malloc(nbytes);
 653         if (buf == NULL) {
 654                 err(udev, "error getting buffer for inotify\n");
 655                 return -1;
 656         }
 657
 658         nbytes = read(pfd[FD_INOTIFY].fd, buf, nbytes);
 659
 660         for (pos = 0; pos < nbytes; pos += sizeof(struct inotify_event) + ev->len) {
 661                 struct udev_device *dev;
 662
 663                 ev = (struct inotify_event *)(buf + pos);
 664                 if (ev->len) {
 665                         const char *s;
 666
 667                         info(udev, "inotify event: %x for %s\n", ev->mask, ev->name);
 668                         s = strstr(ev->name, ".rules");
 669                         if (s == NULL)
 670                                 continue;
 671                         if (strlen(s) != strlen(".rules"))
 672                                 continue;
 673                         reload_config = true;
 674                         continue;
 675                 }
 676
 677                 dev = udev_watch_lookup(udev, ev->wd);
 678                 if (dev != NULL) {
 679                         info(udev, "inotify event: %x for %s\n", ev->mask, udev_device_get_devnode(dev));
 680                         if (ev->mask & IN_CLOSE_WRITE) {
 681                                 char filename[UTIL_PATH_SIZE];
 682                                 int fd;
 683
 684                                 info(udev, "device %s closed, synthesising 'change'\n", udev_device_get_devnode(dev));
 685                                 util_strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
 686                                 fd = open(filename, O_WRONLY);
 687                                 if (fd < 0 || write(fd, "change", 6) < 0)
 688                                         info(udev, "error writing uevent: %m\n");
 689                                 close(fd);
 690                         }
 691                         if (ev->mask & IN_IGNORED)
 692                                 udev_watch_end(udev, dev);
 693
 694                         udev_device_unref(dev);
 695                 }
 696
 697         }
 698
 699         free(buf);
 700         return 0;
 701 }
 702
 703 static void handle_signal(struct udev *udev, int signo)
 704 {
 705         switch (signo) {
 706         case SIGINT:
 707         case SIGTERM:
 708                 udev_exit = true;
 709                 break;
 710         case SIGCHLD:
 711                 while (1) {
 712                         pid_t pid;
 713                         int status;
 714                         struct udev_list_node *loop, *tmp;
 715
 716                         pid = waitpid(-1, &status, WNOHANG);
 717                         if (pid <= 0)
 718                                 break;
 719
 720                         udev_list_node_foreach_safe(loop, tmp, &worker_list) {
 721                                 struct worker *worker = node_to_worker(loop);
 722
 723                                 if (worker->pid != pid)
 724                                         continue;
 725
 726                                 info(udev, "worker [%u] exit\n", pid);
 727                                 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 728                                         err(udev, "worker [%u] unexpectedly returned with status 0x%04x\n", pid, status);
 729                                         if (worker->event != NULL) {
 730                                                 err(udev, "worker [%u] failed while handling '%s'\n", pid, worker->event->devpath);
 731                                                 worker->event->exitcode = -32;
 732                                                 event_queue_delete(worker->event);
 733                                                 /* drop reference from running event */
 734                                                 worker_unref(worker);
 735                                         }
 736                                 }
 737                                 worker_unref(worker);
 738                                 break;
 739                         }
 740                 }
 741                 break;
 742         case SIGHUP:
 743                 reload_config = true;
 744                 break;
 745         }
 746 }
 747
 748 static int copy_dir(struct udev *udev, DIR *dir_from, DIR *dir_to, int maxdepth)
 749 {
 750         struct dirent *dent;
 751
 752         for (dent = readdir(dir_from); dent != NULL; dent = readdir(dir_from)) {
 753                 struct stat stats;
 754
 755                 if (dent->d_name[0] == '.')
 756                         continue;
 757                 if (fstatat(dirfd(dir_from), dent->d_name, &stats, AT_SYMLINK_NOFOLLOW) != 0)
 758                         continue;
 759
 760                 if (S_ISBLK(stats.st_mode) || S_ISCHR(stats.st_mode)) {
 761                         udev_selinux_setfscreateconat(udev, dirfd(dir_to), dent->d_name, stats.st_mode & 0777);
 762                         if (mknodat(dirfd(dir_to), dent->d_name, stats.st_mode, stats.st_rdev) == 0) {
 763                                 fchmodat(dirfd(dir_to), dent->d_name, stats.st_mode & 0777, 0);
 764                                 fchownat(dirfd(dir_to), dent->d_name, stats.st_uid, stats.st_gid, 0);
 765                         } else {
 766                                 utimensat(dirfd(dir_to), dent->d_name, NULL, 0);
 767                         }
 768                         udev_selinux_resetfscreatecon(udev);
 769                 } else if (S_ISLNK(stats.st_mode)) {
 770                         char target[UTIL_PATH_SIZE];
 771                         ssize_t len;
 772
 773                         len = readlinkat(dirfd(dir_from), dent->d_name, target, sizeof(target));
 774                         if (len <= 0 || len == (ssize_t)sizeof(target))
 775                                 continue;
 776                         target[len] = '\0';
 777                         udev_selinux_setfscreateconat(udev, dirfd(dir_to), dent->d_name, S_IFLNK);
 778                         if (symlinkat(target, dirfd(dir_to), dent->d_name) < 0 && errno == EEXIST)
 779                                 utimensat(dirfd(dir_to), dent->d_name, NULL, AT_SYMLINK_NOFOLLOW);
 780                         udev_selinux_resetfscreatecon(udev);
 781                 } else if (S_ISDIR(stats.st_mode)) {
 782                         DIR *dir2_from, *dir2_to;
 783
 784                         if (maxdepth == 0)
 785                                 continue;
 786
 787                         udev_selinux_setfscreateconat(udev, dirfd(dir_to), dent->d_name, S_IFDIR|0755);
 788                         mkdirat(dirfd(dir_to), dent->d_name, 0755);
 789                         udev_selinux_resetfscreatecon(udev);
 790
 791                         dir2_to = fdopendir(openat(dirfd(dir_to), dent->d_name, O_RDONLY|O_NONBLOCK|O_DIRECTORY|O_CLOEXEC));
 792                         if (dir2_to == NULL)
 793                                 continue;
 794
 795                         dir2_from = fdopendir(openat(dirfd(dir_from), dent->d_name, O_RDONLY|O_NONBLOCK|O_DIRECTORY|O_CLOEXEC));
 796                         if (dir2_from == NULL) {
 797                                 closedir(dir2_to);
 798                                 continue;
 799                         }
 800
 801                         copy_dir(udev, dir2_from, dir2_to, maxdepth-1);
 802
 803                         closedir(dir2_to);
 804                         closedir(dir2_from);
 805                 }
 806         }
 807
 808         return 0;
 809 }
 810
 811 static void prepare_dev(struct udev *udev)
 812 {
 813         struct stdlinks {
 814                 const char *link;
 815                 const char *target;
 816         };
 817         static const struct stdlinks stdlinks[] = {
 818                 { "core", "/proc/kcore" },
 819                 { "fd", "/proc/fd" },
 820                 { "stdin", "/proc/self/fd/0" },
 821                 { "stdout", "/proc/self/fd/1" },
 822                 { "stderr", "/proc/self/fd/2" },
 823         };
 824         unsigned int i;
 825         DIR *dir_from, *dir_to;
 826
 827         dir_to = opendir(udev_get_dev_path(udev));
 828         if (dir_to == NULL)
 829                 return;
 830
 831         /* create standard symlinks to /proc */
 832         for (i = 0; i < ARRAY_SIZE(stdlinks); i++) {
 833                 udev_selinux_setfscreateconat(udev, dirfd(dir_to), stdlinks[i].link, S_IFLNK);
 834                 if (symlinkat(stdlinks[i].target, dirfd(dir_to), stdlinks[i].link) < 0 && errno == EEXIST)
 835                         utimensat(dirfd(dir_to), stdlinks[i].link, NULL, AT_SYMLINK_NOFOLLOW);
 836                 udev_selinux_resetfscreatecon(udev);
 837         }
 838
 839         /* copy content from /lib/udev/devices to /dev */
 840         dir_from = opendir(LIBEXECDIR "/devices");
 841         if (dir_from != NULL) {
 842                 copy_dir(udev, dir_from, dir_to, 8);
 843                 closedir(dir_from);
 844         }
 845
 846         closedir(dir_to);
 847 }
 848
 849 static int mem_size_mb(void)
 850 {
 851         FILE *f;
 852         char buf[4096];
 853         long int memsize = -1;
 854
 855         f = fopen("/proc/meminfo", "r");
 856         if (f == NULL)
 857                 return -1;
 858
 859         while (fgets(buf, sizeof(buf), f) != NULL) {
 860                 long int value;
 861
 862                 if (sscanf(buf, "MemTotal: %ld kB", &value) == 1) {
 863                         memsize = value / 1024;
 864                         break;
 865                 }
 866         }
 867
 868         fclose(f);
 869         return memsize;
 870 }
 871
 872 int main(int argc, char *argv[])
 873 {
 874         struct udev *udev;
 875         int fd;
 876         FILE *f;
 877         sigset_t mask;
 878         const char *value;
 879         int daemonize = false;
 880         int resolve_names = 1;
 881         static const struct option options[] = {
 882                 { "daemon", no_argument, NULL, 'd' },
 883                 { "debug-trace", no_argument, NULL, 't' },
 884                 { "debug", no_argument, NULL, 'D' },
 885                 { "help", no_argument, NULL, 'h' },
 886                 { "version", no_argument, NULL, 'V' },
 887                 { "resolve-names", required_argument, NULL, 'N' },
 888                 {}
 889         };
 890         int rc = 1;
 891
 892         udev = udev_new();
 893         if (udev == NULL)
 894                 goto exit;
 895
 896         udev_log_init("udevd");
 897         udev_set_log_fn(udev, log_fn);
 898         info(udev, "version %s\n", VERSION);
 899         udev_selinux_init(udev);
 900
 901         while (1) {
 902                 int option;
 903
 904                 option = getopt_long(argc, argv, "dDthV", options, NULL);
 905                 if (option == -1)
 906                         break;
 907
 908                 switch (option) {
 909                 case 'd':
 910                         daemonize = true;
 911                         break;
 912                 case 't':
 913                         debug_trace = true;
 914                         break;
 915                 case 'D':
 916                         debug = true;
 917                         if (udev_get_log_priority(udev) < LOG_INFO)
 918                                 udev_set_log_priority(udev, LOG_INFO);
 919                         break;
 920                 case 'N':
 921                         if (strcmp (optarg, "early") == 0) {
 922                                 resolve_names = 1;
 923                         } else if (strcmp (optarg, "late") == 0) {
 924                                 resolve_names = 0;
 925                         } else if (strcmp (optarg, "never") == 0) {
 926                                 resolve_names = -1;
 927                         } else {
 928                                 fprintf(stderr, "resolve-names must be early, late or never\n");
 929                                 err(udev, "resolve-names must be early, late or never\n");
 930                                 goto exit;
 931                         }
 932                         break;
 933                 case 'h':
 934                         printf("Usage: udevd [--help] [--daemon] [--debug-trace] [--debug] "
 935                                "[--resolve-names=early|late|never] [--version]\n");
 936                         goto exit;
 937                 case 'V':
 938                         printf("%s\n", VERSION);
 939                         goto exit;
 940                 default:
 941                         goto exit;
 942                 }
 943         }
 944
 945         if (getuid() != 0) {
 946                 fprintf(stderr, "root privileges required\n");
 947                 err(udev, "root privileges required\n");
 948                 goto exit;
 949         }
 950
 951         /* make sure std{in,out,err} fd's are in a sane state */
 952         fd = open("/dev/null", O_RDWR);
 953         if (fd < 0) {
 954                 fprintf(stderr, "cannot open /dev/null\n");
 955                 err(udev, "cannot open /dev/null\n");
 956         }
 957         if (write(STDOUT_FILENO, 0, 0) < 0)
 958                 dup2(fd, STDOUT_FILENO);
 959         if (write(STDERR_FILENO, 0, 0) < 0)
 960                 dup2(fd, STDERR_FILENO);
 961
 962         prepare_dev(udev);
 963
 964         /* init control socket, bind() ensures, that only one udevd instance is running */
 965         udev_ctrl = udev_ctrl_new_from_socket(udev, UDEV_CTRL_SOCK_PATH);
 966         if (udev_ctrl == NULL) {
 967                 fprintf(stderr, "error initializing control socket");
 968                 err(udev, "error initializing udevd socket");
 969                 rc = 1;
 970                 goto exit;
 971         }
 972         if (udev_ctrl_enable_receiving(udev_ctrl) < 0) {
 973                 fprintf(stderr, "error binding control socket, seems udevd is already running\n");
 974                 err(udev, "error binding control socket, seems udevd is already running\n");
 975                 rc = 1;
 976                 goto exit;
 977         }
 978         pfd[FD_CONTROL].fd = udev_ctrl_get_fd(udev_ctrl);
 979
 980         monitor = udev_monitor_new_from_netlink(udev, "kernel");
 981         if (monitor == NULL || udev_monitor_enable_receiving(monitor) < 0) {
 982                 fprintf(stderr, "error initializing netlink socket\n");
 983                 err(udev, "error initializing netlink socket\n");
 984                 rc = 3;
 985                 goto exit;
 986         }
 987         udev_monitor_set_receive_buffer_size(monitor, 128*1024*1024);
 988         pfd[FD_NETLINK].fd = udev_monitor_get_fd(monitor);
 989
 990         pfd[FD_INOTIFY].fd = udev_watch_init(udev);
 991         if (pfd[FD_INOTIFY].fd < 0) {
 992                 fprintf(stderr, "error initializing inotify\n");
 993                 err(udev, "error initializing inotify\n");
 994                 rc = 4;
 995                 goto exit;
 996         }
 997
 998         if (udev_get_rules_path(udev) != NULL) {
 999                 inotify_add_watch(pfd[FD_INOTIFY].fd, udev_get_rules_path(udev),
1000                                   IN_DELETE | IN_MOVE | IN_CLOSE_WRITE);
1001         } else {
1002                 char filename[UTIL_PATH_SIZE];
1003                 struct stat statbuf;
1004
1005                 inotify_add_watch(pfd[FD_INOTIFY].fd, LIBEXECDIR "/rules.d",
1006                                   IN_DELETE | IN_MOVE | IN_CLOSE_WRITE);
1007                 inotify_add_watch(pfd[FD_INOTIFY].fd, SYSCONFDIR "/udev/rules.d",
1008                                   IN_DELETE | IN_MOVE | IN_CLOSE_WRITE);
1009
1010                 /* watch dynamic rules directory */
1011                 util_strscpyl(filename, sizeof(filename), udev_get_dev_path(udev), "/.udev/rules.d", NULL);
1012                 if (stat(filename, &statbuf) != 0) {
1013                         util_create_path(udev, filename);
1014                         udev_selinux_setfscreatecon(udev, filename, S_IFDIR|0755);
1015                         mkdir(filename, 0755);
1016                         udev_selinux_resetfscreatecon(udev);
1017                 }
1018                 inotify_add_watch(pfd[FD_INOTIFY].fd, filename,
1019                                   IN_DELETE | IN_MOVE | IN_CLOSE_WRITE);
1020         }
1021         udev_watch_restore(udev);
1022
1023         /* block and listen to all signals on signalfd */
1024         sigfillset(&mask);
1025         sigprocmask(SIG_SETMASK, &mask, &orig_sigmask);
1026         pfd[FD_SIGNAL].fd = signalfd(-1, &mask, 0);
1027         if (pfd[FD_SIGNAL].fd < 0) {
1028                 fprintf(stderr, "error getting signalfd\n");
1029                 err(udev, "error getting signalfd\n");
1030                 rc = 5;
1031                 goto exit;
1032         }
1033
1034         /* unnamed socket from workers to the main daemon */
1035         if (socketpair(AF_LOCAL, SOCK_DGRAM|SOCK_CLOEXEC, 0, worker_watch) < 0) {
1036                 fprintf(stderr, "error getting socketpair\n");
1037                 err(udev, "error getting socketpair\n");
1038                 rc = 6;
1039                 goto exit;
1040         }
1041         pfd[FD_WORKER].fd = worker_watch[READ_END];
1042
1043         rules = udev_rules_new(udev, resolve_names);
1044         if (rules == NULL) {
1045                 err(udev, "error reading rules\n");
1046                 goto exit;
1047         }
1048
1049         udev_queue_export = udev_queue_export_new(udev);
1050         if (udev_queue_export == NULL) {
1051                 err(udev, "error creating queue file\n");
1052                 goto exit;
1053         }
1054
1055         if (daemonize) {
1056                 pid_t pid;
1057
1058                 pid = fork();
1059                 switch (pid) {
1060                 case 0:
1061                         break;
1062                 case -1:
1063                         err(udev, "fork of daemon failed: %m\n");
1064                         rc = 4;
1065                         goto exit;
1066                 default:
1067                         rc = 0;
1068                         goto exit;
1069                 }
1070         }
1071
1072         f = fopen("/dev/kmsg", "w");
1073         if (f != NULL) {
1074                 fprintf(f, "<6>udev: starting version " VERSION "\n");
1075                 fclose(f);
1076         }
1077
1078         /* redirect std{out,err} */
1079         if (!debug && !debug_trace) {
1080                 dup2(fd, STDIN_FILENO);
1081                 dup2(fd, STDOUT_FILENO);
1082                 dup2(fd, STDERR_FILENO);
1083         }
1084         if (fd > STDERR_FILENO)
1085                 close(fd);
1086
1087         /* set scheduling priority for the daemon */
1088         setpriority(PRIO_PROCESS, 0, UDEVD_PRIORITY);
1089
1090         chdir("/");
1091         umask(022);
1092         setsid();
1093
1094         /* OOM_DISABLE == -17 */
1095         fd = open("/proc/self/oom_adj", O_RDWR);
1096         if (fd < 0) {
1097                 err(udev, "error disabling OOM: %m\n");
1098         } else {
1099                 write(fd, "-17", 3);
1100                 close(fd);
1101         }
1102
1103         /* in trace mode run one event after the other */
1104         if (debug_trace) {
1105                 max_childs = 1;
1106         } else {
1107                 int memsize = mem_size_mb();
1108
1109                 if (memsize > 0)
1110                         max_childs = 128 + (memsize / 8);
1111                 else
1112                         max_childs = 128;
1113         }
1114
1115         /* possibly overwrite maximum limit of executed events */
1116         value = getenv("UDEVD_MAX_CHILDS");
1117         if (value)
1118                 max_childs = strtoul(value, NULL, 10);
1119         info(udev, "initialize max_childs to %u\n", max_childs);
1120
1121         udev_list_init(&event_list);
1122         udev_list_init(&worker_list);
1123
1124         while (!udev_exit) {
1125                 int fdcount;
1126                 int timeout;
1127
1128                 /* set timeout to kill idle workers */
1129                 if (udev_list_is_empty(&event_list) && childs > 2)
1130                         timeout = 3 * 1000;
1131                 else
1132                         timeout = -1;
1133                 /* wait for events */
1134                 fdcount = poll(pfd, ARRAY_SIZE(pfd), timeout);
1135                 if (fdcount < 0)
1136                         continue;
1137
1138                 /* timeout - kill idle workers */
1139                 if (fdcount == 0)
1140                         worker_kill(udev, 2);
1141
1142                 /* event has finished */
1143                 if (pfd[FD_WORKER].revents & POLLIN)
1144                         worker_returned();
1145
1146                 /* get kernel uevent */
1147                 if (pfd[FD_NETLINK].revents & POLLIN) {
1148                         struct udev_device *dev;
1149
1150                         dev = udev_monitor_receive_device(monitor);
1151                         if (dev != NULL)
1152                                 event_queue_insert(dev);
1153                         else
1154                                 udev_device_unref(dev);
1155                 }
1156
1157                 /* start new events */
1158                 if (!udev_list_is_empty(&event_list) && !stop_exec_queue)
1159                         events_start(udev);
1160
1161                 /* get signal */
1162                 if (pfd[FD_SIGNAL].revents & POLLIN) {
1163                         struct signalfd_siginfo fdsi;
1164                         ssize_t size;
1165
1166                         size = read(pfd[FD_SIGNAL].fd, &fdsi, sizeof(struct signalfd_siginfo));
1167                         if (size == sizeof(struct signalfd_siginfo))
1168                                 handle_signal(udev, fdsi.ssi_signo);
1169                 }
1170
1171                 /* device node and rules directory inotify watch */
1172                 if (pfd[FD_INOTIFY].revents & POLLIN)
1173                         handle_inotify(udev);
1174
1175                 /*
1176                  * get control message
1177                  *
1178                  * This needs to be after the inotify handling, to make sure,
1179                  * that the settle signal is send back after the possibly generated
1180                  * "change" events by the inotify device node watch.
1181                  */
1182                 if (pfd[FD_CONTROL].revents & POLLIN)
1183                         handle_ctrl_msg(udev_ctrl);
1184
1185                 /* rules changed, set by inotify or a HUP signal */
1186                 if (reload_config) {
1187                         struct udev_rules *rules_new;
1188
1189                         worker_kill(udev, 0);
1190                         rules_new = udev_rules_new(udev, resolve_names);
1191                         if (rules_new != NULL) {
1192                                 udev_rules_unref(rules);
1193                                 rules = rules_new;
1194                         }
1195                         reload_config = 0;
1196                 }
1197         }
1198
1199         udev_queue_export_cleanup(udev_queue_export);
1200         rc = 0;
1201 exit:
1202         udev_queue_export_unref(udev_queue_export);
1203         udev_rules_unref(rules);
1204         udev_ctrl_unref(udev_ctrl);
1205         if (pfd[FD_SIGNAL].fd >= 0)
1206                 close(pfd[FD_SIGNAL].fd);
1207         if (worker_watch[READ_END] >= 0)
1208                 close(worker_watch[READ_END]);
1209         if (worker_watch[WRITE_END] >= 0)
1210                 close(worker_watch[WRITE_END]);
1211         udev_monitor_unref(monitor);
1212         udev_selinux_exit(udev);
1213         udev_unref(udev);
1214         udev_log_close();
1215         return rc;
1216 }