src/udev/udevd.c

   1 /*
   2  * Copyright (C) 2004-2012 Kay Sievers <kay@vrfy.org>
   3  * Copyright (C) 2004 Chris Friesen <chris_friesen@sympatico.ca>
   4  * Copyright (C) 2009 Canonical Ltd.
   5  * Copyright (C) 2009 Scott James Remnant <scott@netsplit.com>
   6  *
   7  * This program is free software: you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License as published by
   9  * the Free Software Foundation, either version 2 of the License, or
  10  * (at your option) any later version.
  11  *
  12  * This program is distributed in the hope that it will be useful,
  13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  15  * GNU General Public License for more details.
  16  *
  17  * You should have received a copy of the GNU General Public License
  18  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
  19  */
  20
  21 #include <stddef.h>
  22 #include <signal.h>
  23 #include <unistd.h>
  24 #include <errno.h>
  25 #include <stdio.h>
  26 #include <stdlib.h>
  27 #include <stdbool.h>
  28 #include <string.h>
  29 #include <fcntl.h>
  30 #include <getopt.h>
  31 #include <sys/file.h>
  32 #include <sys/time.h>
  33 #include <sys/prctl.h>
  34 #include <sys/socket.h>
  35 #include <sys/signalfd.h>
  36 #include <sys/epoll.h>
  37 #include <sys/mount.h>
  38 #include <sys/wait.h>
  39 #include <sys/stat.h>
  40 #include <sys/ioctl.h>
  41 #include <sys/inotify.h>
  42
  43 #include "sd-daemon.h"
  44 #include "rtnl-util.h"
  45 #include "cgroup-util.h"
  46 #include "dev-setup.h"
  47 #include "fileio.h"
  48 #include "selinux-util.h"
  49 #include "udev.h"
  50 #include "udev-util.h"
  51
  52 static struct udev_rules *rules;
  53 static struct udev_ctrl *udev_ctrl;
  54 static struct udev_monitor *monitor;
  55 static int worker_watch[2] = { -1, -1 };
  56 static int fd_signal = -1;
  57 static int fd_ep = -1;
  58 static int fd_inotify = -1;
  59 static bool stop_exec_queue;
  60 static bool reload;
  61 static int children;
  62 static bool arg_debug = false;
  63 static int arg_daemonize = false;
  64 static int arg_resolve_names = 1;
  65 static int arg_children_max;
  66 static int arg_exec_delay;
  67 static usec_t arg_event_timeout_usec = 180 * USEC_PER_SEC;
  68 static usec_t arg_event_timeout_warn_usec = 180 * USEC_PER_SEC / 3;
  69 static sigset_t sigmask_orig;
  70 static UDEV_LIST(event_list);
  71 static UDEV_LIST(worker_list);
  72 static char *udev_cgroup;
  73 static struct udev_list properties_list;
  74 static bool udev_exit;
  75
  76 enum event_state {
  77         EVENT_UNDEF,
  78         EVENT_QUEUED,
  79         EVENT_RUNNING,
  80 };
  81
  82 struct event {
  83         struct udev_list_node node;
  84         struct udev *udev;
  85         struct udev_device *dev;
  86         struct udev_device *dev_kernel;
  87         enum event_state state;
  88         int exitcode;
  89         unsigned long long int delaying_seqnum;
  90         unsigned long long int seqnum;
  91         const char *devpath;
  92         size_t devpath_len;
  93         const char *devpath_old;
  94         dev_t devnum;
  95         int ifindex;
  96         bool is_block;
  97 };
  98
  99 static inline struct event *node_to_event(struct udev_list_node *node) {
 100         return container_of(node, struct event, node);
 101 }
 102
 103 static void event_queue_cleanup(struct udev *udev, enum event_state type);
 104
 105 enum worker_state {
 106         WORKER_UNDEF,
 107         WORKER_RUNNING,
 108         WORKER_IDLE,
 109         WORKER_KILLED,
 110 };
 111
 112 struct worker {
 113         struct udev_list_node node;
 114         struct udev *udev;
 115         int refcount;
 116         pid_t pid;
 117         struct udev_monitor *monitor;
 118         enum worker_state state;
 119         struct event *event;
 120         usec_t event_start_usec;
 121         bool event_warned;
 122 };
 123
 124 /* passed from worker to main process */
 125 struct worker_message {
 126         pid_t pid;
 127         int exitcode;
 128 };
 129
 130 static inline struct worker *node_to_worker(struct udev_list_node *node) {
 131         return container_of(node, struct worker, node);
 132 }
 133
 134 static void event_queue_delete(struct event *event) {
 135         udev_list_node_remove(&event->node);
 136         udev_device_unref(event->dev);
 137         udev_device_unref(event->dev_kernel);
 138         free(event);
 139 }
 140
 141 static struct worker *worker_ref(struct worker *worker) {
 142         worker->refcount++;
 143         return worker;
 144 }
 145
 146 static void worker_cleanup(struct worker *worker) {
 147         udev_list_node_remove(&worker->node);
 148         udev_monitor_unref(worker->monitor);
 149         children--;
 150         free(worker);
 151 }
 152
 153 static void worker_unref(struct worker *worker) {
 154         worker->refcount--;
 155         if (worker->refcount > 0)
 156                 return;
 157         log_debug("worker ["PID_FMT"] cleaned up", worker->pid);
 158         worker_cleanup(worker);
 159 }
 160
 161 static void worker_list_cleanup(struct udev *udev) {
 162         struct udev_list_node *loop, *tmp;
 163
 164         udev_list_node_foreach_safe(loop, tmp, &worker_list) {
 165                 struct worker *worker = node_to_worker(loop);
 166
 167                 worker_cleanup(worker);
 168         }
 169 }
 170
 171 static void worker_new(struct event *event) {
 172         struct udev *udev = event->udev;
 173         struct worker *worker;
 174         struct udev_monitor *worker_monitor;
 175         pid_t pid;
 176
 177         /* listen for new events */
 178         worker_monitor = udev_monitor_new_from_netlink(udev, NULL);
 179         if (worker_monitor == NULL)
 180                 return;
 181         /* allow the main daemon netlink address to send devices to the worker */
 182         udev_monitor_allow_unicast_sender(worker_monitor, monitor);
 183         udev_monitor_enable_receiving(worker_monitor);
 184
 185         worker = new0(struct worker, 1);
 186         if (worker == NULL) {
 187                 udev_monitor_unref(worker_monitor);
 188                 return;
 189         }
 190         /* worker + event reference */
 191         worker->refcount = 2;
 192         worker->udev = udev;
 193
 194         pid = fork();
 195         switch (pid) {
 196         case 0: {
 197                 struct udev_device *dev = NULL;
 198                 int fd_monitor;
 199                 _cleanup_rtnl_unref_ sd_rtnl *rtnl = NULL;
 200                 struct epoll_event ep_signal, ep_monitor;
 201                 sigset_t mask;
 202                 int rc = EXIT_SUCCESS;
 203
 204                 /* take initial device from queue */
 205                 dev = event->dev;
 206                 event->dev = NULL;
 207
 208                 free(worker);
 209                 worker_list_cleanup(udev);
 210                 event_queue_cleanup(udev, EVENT_UNDEF);
 211                 udev_monitor_unref(monitor);
 212                 udev_ctrl_unref(udev_ctrl);
 213                 close(fd_signal);
 214                 close(fd_ep);
 215                 close(worker_watch[READ_END]);
 216
 217                 sigfillset(&mask);
 218                 fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
 219                 if (fd_signal < 0) {
 220                         log_error_errno(errno, "error creating signalfd %m");
 221                         rc = 2;
 222                         goto out;
 223                 }
 224
 225                 fd_ep = epoll_create1(EPOLL_CLOEXEC);
 226                 if (fd_ep < 0) {
 227                         log_error_errno(errno, "error creating epoll fd: %m");
 228                         rc = 3;
 229                         goto out;
 230                 }
 231
 232                 memzero(&ep_signal, sizeof(struct epoll_event));
 233                 ep_signal.events = EPOLLIN;
 234                 ep_signal.data.fd = fd_signal;
 235
 236                 fd_monitor = udev_monitor_get_fd(worker_monitor);
 237                 memzero(&ep_monitor, sizeof(struct epoll_event));
 238                 ep_monitor.events = EPOLLIN;
 239                 ep_monitor.data.fd = fd_monitor;
 240
 241                 if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
 242                     epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_monitor, &ep_monitor) < 0) {
 243                         log_error_errno(errno, "fail to add fds to epoll: %m");
 244                         rc = 4;
 245                         goto out;
 246                 }
 247
 248                 /* request TERM signal if parent exits */
 249                 prctl(PR_SET_PDEATHSIG, SIGTERM);
 250
 251                 /* reset OOM score, we only protect the main daemon */
 252                 write_string_file("/proc/self/oom_score_adj", "0");
 253
 254                 for (;;) {
 255                         struct udev_event *udev_event;
 256                         struct worker_message msg;
 257                         int fd_lock = -1;
 258                         int err = 0;
 259
 260                         log_debug("seq %llu running", udev_device_get_seqnum(dev));
 261                         udev_event = udev_event_new(dev);
 262                         if (udev_event == NULL) {
 263                                 rc = 5;
 264                                 goto out;
 265                         }
 266
 267                         /* needed for SIGCHLD/SIGTERM in spawn() */
 268                         udev_event->fd_signal = fd_signal;
 269
 270                         if (arg_exec_delay > 0)
 271                                 udev_event->exec_delay = arg_exec_delay;
 272
 273                         /*
 274                          * Take a shared lock on the device node; this establishes
 275                          * a concept of device "ownership" to serialize device
 276                          * access. External processes holding an exclusive lock will
 277                          * cause udev to skip the event handling; in the case udev
 278                          * acquired the lock, the external process can block until
 279                          * udev has finished its event handling.
 280                          */
 281                         if (!streq_ptr(udev_device_get_action(dev), "remove") &&
 282                             streq_ptr("block", udev_device_get_subsystem(dev)) &&
 283                             !startswith(udev_device_get_sysname(dev), "dm-") &&
 284                             !startswith(udev_device_get_sysname(dev), "md")) {
 285                                 struct udev_device *d = dev;
 286
 287                                 if (streq_ptr("partition", udev_device_get_devtype(d)))
 288                                         d = udev_device_get_parent(d);
 289
 290                                 if (d) {
 291                                         fd_lock = open(udev_device_get_devnode(d), O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
 292                                         if (fd_lock >= 0 && flock(fd_lock, LOCK_SH|LOCK_NB) < 0) {
 293                                                 log_debug_errno(errno, "Unable to flock(%s), skipping event handling: %m", udev_device_get_devnode(d));
 294                                                 err = -EWOULDBLOCK;
 295                                                 fd_lock = safe_close(fd_lock);
 296                                                 goto skip;
 297                                         }
 298                                 }
 299                         }
 300
 301                         /* needed for renaming netifs */
 302                         udev_event->rtnl = rtnl;
 303
 304                         /* apply rules, create node, symlinks */
 305                         udev_event_execute_rules(udev_event,
 306                                                  arg_event_timeout_usec, arg_event_timeout_warn_usec,
 307                                                  &properties_list,
 308                                                  rules,
 309                                                  &sigmask_orig);
 310
 311                         udev_event_execute_run(udev_event,
 312                                                arg_event_timeout_usec, arg_event_timeout_warn_usec,
 313                                                &sigmask_orig);
 314
 315                         if (udev_event->rtnl)
 316                                 /* in case rtnl was initialized */
 317                                 rtnl = sd_rtnl_ref(udev_event->rtnl);
 318
 319                         /* apply/restore inotify watch */
 320                         if (udev_event->inotify_watch) {
 321                                 udev_watch_begin(udev, dev);
 322                                 udev_device_update_db(dev);
 323                         }
 324
 325                         safe_close(fd_lock);
 326
 327                         /* send processed event back to libudev listeners */
 328                         udev_monitor_send_device(worker_monitor, NULL, dev);
 329
 330 skip:
 331                         /* send udevd the result of the event execution */
 332                         memzero(&msg, sizeof(struct worker_message));
 333                         msg.exitcode = err;
 334                         msg.pid = getpid();
 335                         send(worker_watch[WRITE_END], &msg, sizeof(struct worker_message), 0);
 336
 337                         log_debug("seq %llu processed with %i", udev_device_get_seqnum(dev), err);
 338
 339                         udev_device_unref(dev);
 340                         dev = NULL;
 341
 342                         if (udev_event->sigterm) {
 343                                 udev_event_unref(udev_event);
 344                                 goto out;
 345                         }
 346
 347                         udev_event_unref(udev_event);
 348
 349                         /* wait for more device messages from main udevd, or term signal */
 350                         while (dev == NULL) {
 351                                 struct epoll_event ev[4];
 352                                 int fdcount;
 353                                 int i;
 354
 355                                 fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), -1);
 356                                 if (fdcount < 0) {
 357                                         if (errno == EINTR)
 358                                                 continue;
 359                                         log_error_errno(errno, "failed to poll: %m");
 360                                         goto out;
 361                                 }
 362
 363                                 for (i = 0; i < fdcount; i++) {
 364                                         if (ev[i].data.fd == fd_monitor && ev[i].events & EPOLLIN) {
 365                                                 dev = udev_monitor_receive_device(worker_monitor);
 366                                                 break;
 367                                         } else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN) {
 368                                                 struct signalfd_siginfo fdsi;
 369                                                 ssize_t size;
 370
 371                                                 size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
 372                                                 if (size != sizeof(struct signalfd_siginfo))
 373                                                         continue;
 374                                                 switch (fdsi.ssi_signo) {
 375                                                 case SIGTERM:
 376                                                         goto out;
 377                                                 }
 378                                         }
 379                                 }
 380                         }
 381                 }
 382 out:
 383                 udev_device_unref(dev);
 384                 safe_close(fd_signal);
 385                 safe_close(fd_ep);
 386                 close(fd_inotify);
 387                 close(worker_watch[WRITE_END]);
 388                 udev_rules_unref(rules);
 389                 udev_builtin_exit(udev);
 390                 udev_monitor_unref(worker_monitor);
 391                 udev_unref(udev);
 392                 log_close();
 393                 exit(rc);
 394         }
 395         case -1:
 396                 udev_monitor_unref(worker_monitor);
 397                 event->state = EVENT_QUEUED;
 398                 free(worker);
 399                 log_error_errno(errno, "fork of child failed: %m");
 400                 break;
 401         default:
 402                 /* close monitor, but keep address around */
 403                 udev_monitor_disconnect(worker_monitor);
 404                 worker->monitor = worker_monitor;
 405                 worker->pid = pid;
 406                 worker->state = WORKER_RUNNING;
 407                 worker->event_start_usec = now(CLOCK_MONOTONIC);
 408                 worker->event_warned = false;
 409                 worker->event = event;
 410                 event->state = EVENT_RUNNING;
 411                 udev_list_node_append(&worker->node, &worker_list);
 412                 children++;
 413                 log_debug("seq %llu forked new worker ["PID_FMT"]", udev_device_get_seqnum(event->dev), pid);
 414                 break;
 415         }
 416 }
 417
 418 static void event_run(struct event *event) {
 419         struct udev_list_node *loop;
 420
 421         udev_list_node_foreach(loop, &worker_list) {
 422                 struct worker *worker = node_to_worker(loop);
 423                 ssize_t count;
 424
 425                 if (worker->state != WORKER_IDLE)
 426                         continue;
 427
 428                 count = udev_monitor_send_device(monitor, worker->monitor, event->dev);
 429                 if (count < 0) {
 430                         log_error_errno(errno, "worker ["PID_FMT"] did not accept message %zi (%m), kill it",
 431                                         worker->pid, count);
 432                         kill(worker->pid, SIGKILL);
 433                         worker->state = WORKER_KILLED;
 434                         continue;
 435                 }
 436                 worker_ref(worker);
 437                 worker->event = event;
 438                 worker->state = WORKER_RUNNING;
 439                 worker->event_start_usec = now(CLOCK_MONOTONIC);
 440                 worker->event_warned = false;
 441                 event->state = EVENT_RUNNING;
 442                 return;
 443         }
 444
 445         if (children >= arg_children_max) {
 446                 if (arg_children_max > 1)
 447                         log_debug("maximum number (%i) of children reached", children);
 448                 return;
 449         }
 450
 451         /* start new worker and pass initial device */
 452         worker_new(event);
 453 }
 454
 455 static int event_queue_insert(struct udev_device *dev) {
 456         struct event *event;
 457
 458         event = new0(struct event, 1);
 459         if (event == NULL)
 460                 return -1;
 461
 462         event->udev = udev_device_get_udev(dev);
 463         event->dev = dev;
 464         event->dev_kernel = udev_device_shallow_clone(dev);
 465         udev_device_copy_properties(event->dev_kernel, dev);
 466         event->seqnum = udev_device_get_seqnum(dev);
 467         event->devpath = udev_device_get_devpath(dev);
 468         event->devpath_len = strlen(event->devpath);
 469         event->devpath_old = udev_device_get_devpath_old(dev);
 470         event->devnum = udev_device_get_devnum(dev);
 471         event->is_block = streq("block", udev_device_get_subsystem(dev));
 472         event->ifindex = udev_device_get_ifindex(dev);
 473
 474         log_debug("seq %llu queued, '%s' '%s'", udev_device_get_seqnum(dev),
 475              udev_device_get_action(dev), udev_device_get_subsystem(dev));
 476
 477         event->state = EVENT_QUEUED;
 478         udev_list_node_append(&event->node, &event_list);
 479         return 0;
 480 }
 481
 482 static void worker_kill(struct udev *udev) {
 483         struct udev_list_node *loop;
 484
 485         udev_list_node_foreach(loop, &worker_list) {
 486                 struct worker *worker = node_to_worker(loop);
 487
 488                 if (worker->state == WORKER_KILLED)
 489                         continue;
 490
 491                 worker->state = WORKER_KILLED;
 492                 kill(worker->pid, SIGTERM);
 493         }
 494 }
 495
 496 /* lookup event for identical, parent, child device */
 497 static bool is_devpath_busy(struct event *event) {
 498         struct udev_list_node *loop;
 499         size_t common;
 500
 501         /* check if queue contains events we depend on */
 502         udev_list_node_foreach(loop, &event_list) {
 503                 struct event *loop_event = node_to_event(loop);
 504
 505                 /* we already found a later event, earlier can not block us, no need to check again */
 506                 if (loop_event->seqnum < event->delaying_seqnum)
 507                         continue;
 508
 509                 /* event we checked earlier still exists, no need to check again */
 510                 if (loop_event->seqnum == event->delaying_seqnum)
 511                         return true;
 512
 513                 /* found ourself, no later event can block us */
 514                 if (loop_event->seqnum >= event->seqnum)
 515                         break;
 516
 517                 /* check major/minor */
 518                 if (major(event->devnum) != 0 && event->devnum == loop_event->devnum && event->is_block == loop_event->is_block)
 519                         return true;
 520
 521                 /* check network device ifindex */
 522                 if (event->ifindex != 0 && event->ifindex == loop_event->ifindex)
 523                         return true;
 524
 525                 /* check our old name */
 526                 if (event->devpath_old != NULL && streq(loop_event->devpath, event->devpath_old)) {
 527                         event->delaying_seqnum = loop_event->seqnum;
 528                         return true;
 529                 }
 530
 531                 /* compare devpath */
 532                 common = MIN(loop_event->devpath_len, event->devpath_len);
 533
 534                 /* one devpath is contained in the other? */
 535                 if (memcmp(loop_event->devpath, event->devpath, common) != 0)
 536                         continue;
 537
 538                 /* identical device event found */
 539                 if (loop_event->devpath_len == event->devpath_len) {
 540                         /* devices names might have changed/swapped in the meantime */
 541                         if (major(event->devnum) != 0 && (event->devnum != loop_event->devnum || event->is_block != loop_event->is_block))
 542                                 continue;
 543                         if (event->ifindex != 0 && event->ifindex != loop_event->ifindex)
 544                                 continue;
 545                         event->delaying_seqnum = loop_event->seqnum;
 546                         return true;
 547                 }
 548
 549                 /* parent device event found */
 550                 if (event->devpath[common] == '/') {
 551                         event->delaying_seqnum = loop_event->seqnum;
 552                         return true;
 553                 }
 554
 555                 /* child device event found */
 556                 if (loop_event->devpath[common] == '/') {
 557                         event->delaying_seqnum = loop_event->seqnum;
 558                         return true;
 559                 }
 560
 561                 /* no matching device */
 562                 continue;
 563         }
 564
 565         return false;
 566 }
 567
 568 static void event_queue_start(struct udev *udev) {
 569         struct udev_list_node *loop;
 570
 571         udev_list_node_foreach(loop, &event_list) {
 572                 struct event *event = node_to_event(loop);
 573
 574                 if (event->state != EVENT_QUEUED)
 575                         continue;
 576
 577                 /* do not start event if parent or child event is still running */
 578                 if (is_devpath_busy(event))
 579                         continue;
 580
 581                 event_run(event);
 582         }
 583 }
 584
 585 static void event_queue_cleanup(struct udev *udev, enum event_state match_type) {
 586         struct udev_list_node *loop, *tmp;
 587
 588         udev_list_node_foreach_safe(loop, tmp, &event_list) {
 589                 struct event *event = node_to_event(loop);
 590
 591                 if (match_type != EVENT_UNDEF && match_type != event->state)
 592                         continue;
 593
 594                 event_queue_delete(event);
 595         }
 596 }
 597
 598 static void worker_returned(int fd_worker) {
 599         for (;;) {
 600                 struct worker_message msg;
 601                 ssize_t size;
 602                 struct udev_list_node *loop;
 603
 604                 size = recv(fd_worker, &msg, sizeof(struct worker_message), MSG_DONTWAIT);
 605                 if (size != sizeof(struct worker_message))
 606                         break;
 607
 608                 /* lookup worker who sent the signal */
 609                 udev_list_node_foreach(loop, &worker_list) {
 610                         struct worker *worker = node_to_worker(loop);
 611
 612                         if (worker->pid != msg.pid)
 613                                 continue;
 614
 615                         /* worker returned */
 616                         if (worker->event) {
 617                                 worker->event->exitcode = msg.exitcode;
 618                                 event_queue_delete(worker->event);
 619                                 worker->event = NULL;
 620                         }
 621                         if (worker->state != WORKER_KILLED)
 622                                 worker->state = WORKER_IDLE;
 623                         worker_unref(worker);
 624                         break;
 625                 }
 626         }
 627 }
 628
 629 /* receive the udevd message from userspace */
 630 static struct udev_ctrl_connection *handle_ctrl_msg(struct udev_ctrl *uctrl) {
 631         struct udev *udev = udev_ctrl_get_udev(uctrl);
 632         struct udev_ctrl_connection *ctrl_conn;
 633         struct udev_ctrl_msg *ctrl_msg = NULL;
 634         const char *str;
 635         int i;
 636
 637         ctrl_conn = udev_ctrl_get_connection(uctrl);
 638         if (ctrl_conn == NULL)
 639                 goto out;
 640
 641         ctrl_msg = udev_ctrl_receive_msg(ctrl_conn);
 642         if (ctrl_msg == NULL)
 643                 goto out;
 644
 645         i = udev_ctrl_get_set_log_level(ctrl_msg);
 646         if (i >= 0) {
 647                 log_debug("udevd message (SET_LOG_LEVEL) received, log_priority=%i", i);
 648                 log_set_max_level(i);
 649                 worker_kill(udev);
 650         }
 651
 652         if (udev_ctrl_get_stop_exec_queue(ctrl_msg) > 0) {
 653                 log_debug("udevd message (STOP_EXEC_QUEUE) received");
 654                 stop_exec_queue = true;
 655         }
 656
 657         if (udev_ctrl_get_start_exec_queue(ctrl_msg) > 0) {
 658                 log_debug("udevd message (START_EXEC_QUEUE) received");
 659                 stop_exec_queue = false;
 660         }
 661
 662         if (udev_ctrl_get_reload(ctrl_msg) > 0) {
 663                 log_debug("udevd message (RELOAD) received");
 664                 reload = true;
 665         }
 666
 667         str = udev_ctrl_get_set_env(ctrl_msg);
 668         if (str != NULL) {
 669                 char *key;
 670
 671                 key = strdup(str);
 672                 if (key != NULL) {
 673                         char *val;
 674
 675                         val = strchr(key, '=');
 676                         if (val != NULL) {
 677                                 val[0] = '\0';
 678                                 val = &val[1];
 679                                 if (val[0] == '\0') {
 680                                         log_debug("udevd message (ENV) received, unset '%s'", key);
 681                                         udev_list_entry_add(&properties_list, key, NULL);
 682                                 } else {
 683                                         log_debug("udevd message (ENV) received, set '%s=%s'", key, val);
 684                                         udev_list_entry_add(&properties_list, key, val);
 685                                 }
 686                         } else {
 687                                 log_error("wrong key format '%s'", key);
 688                         }
 689                         free(key);
 690                 }
 691                 worker_kill(udev);
 692         }
 693
 694         i = udev_ctrl_get_set_children_max(ctrl_msg);
 695         if (i >= 0) {
 696                 log_debug("udevd message (SET_MAX_CHILDREN) received, children_max=%i", i);
 697                 arg_children_max = i;
 698         }
 699
 700         if (udev_ctrl_get_ping(ctrl_msg) > 0)
 701                 log_debug("udevd message (SYNC) received");
 702
 703         if (udev_ctrl_get_exit(ctrl_msg) > 0) {
 704                 log_debug("udevd message (EXIT) received");
 705                 udev_exit = true;
 706                 /* keep reference to block the client until we exit */
 707                 udev_ctrl_connection_ref(ctrl_conn);
 708         }
 709 out:
 710         udev_ctrl_msg_unref(ctrl_msg);
 711         return udev_ctrl_connection_unref(ctrl_conn);
 712 }
 713
 714 static int synthesize_change(struct udev_device *dev) {
 715         char filename[UTIL_PATH_SIZE];
 716         int r;
 717
 718         if (streq_ptr("block", udev_device_get_subsystem(dev)) &&
 719             streq_ptr("disk", udev_device_get_devtype(dev)) &&
 720             !startswith(udev_device_get_sysname(dev), "dm-")) {
 721                 bool part_table_read = false;
 722                 bool has_partitions = false;
 723                 int fd;
 724                 struct udev *udev = udev_device_get_udev(dev);
 725                 _cleanup_udev_enumerate_unref_ struct udev_enumerate *e = NULL;
 726                 struct udev_list_entry *item;
 727
 728                 /*
 729                  * Try to re-read the partition table. This only succeeds if
 730                  * none of the devices is busy. The kernel returns 0 if no
 731                  * partition table is found, and we will not get an event for
 732                  * the disk.
 733                  */
 734                 fd = open(udev_device_get_devnode(dev), O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
 735                 if (fd >= 0) {
 736                         r = flock(fd, LOCK_EX|LOCK_NB);
 737                         if (r >= 0)
 738                                 r = ioctl(fd, BLKRRPART, 0);
 739
 740                         close(fd);
 741                         if (r >= 0)
 742                                 part_table_read = true;
 743                 }
 744
 745                 /* search for partitions */
 746                 e = udev_enumerate_new(udev);
 747                 if (!e)
 748                         return -ENOMEM;
 749
 750                 r = udev_enumerate_add_match_parent(e, dev);
 751                 if (r < 0)
 752                         return r;
 753
 754                 r = udev_enumerate_add_match_subsystem(e, "block");
 755                 if (r < 0)
 756                         return r;
 757
 758                 r = udev_enumerate_scan_devices(e);
 759                 if (r < 0)
 760                         return r;
 761
 762                 udev_list_entry_foreach(item, udev_enumerate_get_list_entry(e)) {
 763                         _cleanup_udev_device_unref_ struct udev_device *d = NULL;
 764
 765                         d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
 766                         if (!d)
 767                                 continue;
 768
 769                         if (!streq_ptr("partition", udev_device_get_devtype(d)))
 770                                 continue;
 771
 772                         has_partitions = true;
 773                         break;
 774                 }
 775
 776                 /*
 777                  * We have partitions and re-read the table, the kernel already sent
 778                  * out a "change" event for the disk, and "remove/add" for all
 779                  * partitions.
 780                  */
 781                 if (part_table_read && has_partitions)
 782                         return 0;
 783
 784                 /*
 785                  * We have partitions but re-reading the partition table did not
 786                  * work, synthesize "change" for the disk and all partitions.
 787                  */
 788                 log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev));
 789                 strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
 790                 write_string_file(filename, "change");
 791
 792                 udev_list_entry_foreach(item, udev_enumerate_get_list_entry(e)) {
 793                         _cleanup_udev_device_unref_ struct udev_device *d = NULL;
 794
 795                         d = udev_device_new_from_syspath(udev, udev_list_entry_get_name(item));
 796                         if (!d)
 797                                 continue;
 798
 799                         if (!streq_ptr("partition", udev_device_get_devtype(d)))
 800                                 continue;
 801
 802                         log_debug("device %s closed, synthesising partition '%s' 'change'",
 803                                   udev_device_get_devnode(dev), udev_device_get_devnode(d));
 804                         strscpyl(filename, sizeof(filename), udev_device_get_syspath(d), "/uevent", NULL);
 805                         write_string_file(filename, "change");
 806                 }
 807
 808                 return 0;
 809         }
 810
 811         log_debug("device %s closed, synthesising 'change'", udev_device_get_devnode(dev));
 812         strscpyl(filename, sizeof(filename), udev_device_get_syspath(dev), "/uevent", NULL);
 813         write_string_file(filename, "change");
 814
 815         return 0;
 816 }
 817
 818 static int handle_inotify(struct udev *udev) {
 819         union inotify_event_buffer buffer;
 820         struct inotify_event *e;
 821         ssize_t l;
 822
 823         l = read(fd_inotify, &buffer, sizeof(buffer));
 824         if (l < 0) {
 825                 if (errno == EAGAIN || errno == EINTR)
 826                         return 0;
 827
 828                 return log_error_errno(errno, "Failed to read inotify fd: %m");
 829         }
 830
 831         FOREACH_INOTIFY_EVENT(e, buffer, l) {
 832                 struct udev_device *dev;
 833
 834                 dev = udev_watch_lookup(udev, e->wd);
 835                 if (!dev)
 836                         continue;
 837
 838                 log_debug("inotify event: %x for %s", e->mask, udev_device_get_devnode(dev));
 839                 if (e->mask & IN_CLOSE_WRITE)
 840                         synthesize_change(dev);
 841                 else if (e->mask & IN_IGNORED)
 842                         udev_watch_end(udev, dev);
 843
 844                 udev_device_unref(dev);
 845         }
 846
 847         return 0;
 848 }
 849
 850 static void handle_signal(struct udev *udev, int signo) {
 851         switch (signo) {
 852         case SIGINT:
 853         case SIGTERM:
 854                 udev_exit = true;
 855                 break;
 856         case SIGCHLD:
 857                 for (;;) {
 858                         pid_t pid;
 859                         int status;
 860                         struct udev_list_node *loop, *tmp;
 861
 862                         pid = waitpid(-1, &status, WNOHANG);
 863                         if (pid <= 0)
 864                                 break;
 865
 866                         udev_list_node_foreach_safe(loop, tmp, &worker_list) {
 867                                 struct worker *worker = node_to_worker(loop);
 868
 869                                 if (worker->pid != pid)
 870                                         continue;
 871                                 log_debug("worker ["PID_FMT"] exit", pid);
 872
 873                                 if (WIFEXITED(status)) {
 874                                         if (WEXITSTATUS(status) != 0)
 875                                                 log_error("worker ["PID_FMT"] exit with return code %i",
 876                                                           pid, WEXITSTATUS(status));
 877                                 } else if (WIFSIGNALED(status)) {
 878                                         log_error("worker ["PID_FMT"] terminated by signal %i (%s)",
 879                                                   pid, WTERMSIG(status), strsignal(WTERMSIG(status)));
 880                                 } else if (WIFSTOPPED(status)) {
 881                                         log_error("worker ["PID_FMT"] stopped", pid);
 882                                 } else if (WIFCONTINUED(status)) {
 883                                         log_error("worker ["PID_FMT"] continued", pid);
 884                                 } else {
 885                                         log_error("worker ["PID_FMT"] exit with status 0x%04x", pid, status);
 886                                 }
 887
 888                                 if (!WIFEXITED(status) || WEXITSTATUS(status) != 0) {
 889                                         if (worker->event) {
 890                                                 log_error("worker ["PID_FMT"] failed while handling '%s'",
 891                                                           pid, worker->event->devpath);
 892                                                 worker->event->exitcode = -32;
 893                                                 /* delete state from disk */
 894                                                 udev_device_delete_db(worker->event->dev);
 895                                                 udev_device_tag_index(worker->event->dev, NULL, false);
 896                                                 /* forward kernel event without ammending it */
 897                                                 udev_monitor_send_device(monitor, NULL, worker->event->dev_kernel);
 898                                                 event_queue_delete(worker->event);
 899
 900                                                 /* drop reference taken for state 'running' */
 901                                                 worker_unref(worker);
 902                                         }
 903                                 }
 904                                 worker_unref(worker);
 905                                 break;
 906                         }
 907                 }
 908                 break;
 909         case SIGHUP:
 910                 reload = true;
 911                 break;
 912         }
 913 }
 914
 915 static void event_queue_update(void) {
 916         int r;
 917
 918         if (!udev_list_node_is_empty(&event_list)) {
 919                 r = touch("/run/udev/queue");
 920                 if (r < 0)
 921                         log_warning_errno(r, "could not touch /run/udev/queue: %m");
 922         } else {
 923                 r = unlink("/run/udev/queue");
 924                 if (r < 0 && errno != ENOENT)
 925                         log_warning("could not unlink /run/udev/queue: %m");
 926         }
 927 }
 928
 929 static int systemd_fds(struct udev *udev, int *rctrl, int *rnetlink) {
 930         int ctrl = -1, netlink = -1;
 931         int fd, n;
 932
 933         n = sd_listen_fds(true);
 934         if (n <= 0)
 935                 return -1;
 936
 937         for (fd = SD_LISTEN_FDS_START; fd < n + SD_LISTEN_FDS_START; fd++) {
 938                 if (sd_is_socket(fd, AF_LOCAL, SOCK_SEQPACKET, -1)) {
 939                         if (ctrl >= 0)
 940                                 return -1;
 941                         ctrl = fd;
 942                         continue;
 943                 }
 944
 945                 if (sd_is_socket(fd, AF_NETLINK, SOCK_RAW, -1)) {
 946                         if (netlink >= 0)
 947                                 return -1;
 948                         netlink = fd;
 949                         continue;
 950                 }
 951
 952                 return -1;
 953         }
 954
 955         if (ctrl < 0 || netlink < 0)
 956                 return -1;
 957
 958         log_debug("ctrl=%i netlink=%i", ctrl, netlink);
 959         *rctrl = ctrl;
 960         *rnetlink = netlink;
 961         return 0;
 962 }
 963
 964 /*
 965  * read the kernel command line, in case we need to get into debug mode
 966  *   udev.log-priority=<level>              syslog priority
 967  *   udev.children-max=<number of workers>  events are fully serialized if set to 1
 968  *   udev.exec-delay=<number of seconds>    delay execution of every executed program
 969  */
 970 static void kernel_cmdline_options(struct udev *udev) {
 971         _cleanup_free_ char *line = NULL;
 972         const char *word, *state;
 973         size_t l;
 974         int r;
 975
 976         r = proc_cmdline(&line);
 977         if (r < 0) {
 978                 log_warning_errno(r, "Failed to read /proc/cmdline, ignoring: %m");
 979                 return;
 980         }
 981
 982         FOREACH_WORD_QUOTED(word, l, line, state) {
 983                 char *s, *opt, *value;
 984
 985                 s = strndup(word, l);
 986                 if (!s)
 987                         break;
 988
 989                 /* accept the same options for the initrd, prefixed with "rd." */
 990                 if (in_initrd() && startswith(s, "rd."))
 991                         opt = s + 3;
 992                 else
 993                         opt = s;
 994
 995                 if ((value = startswith(opt, "udev.log-priority="))) {
 996                         int prio;
 997
 998                         prio = util_log_priority(value);
 999                         log_set_max_level(prio);
1000                 } else if ((value = startswith(opt, "udev.children-max="))) {
1001                         r = safe_atoi(value, &arg_children_max);
1002                         if (r < 0)
1003                                 log_warning("Invalid udev.children-max ignored: %s", value);
1004                 } else if ((value = startswith(opt, "udev.exec-delay="))) {
1005                         r = safe_atoi(value, &arg_exec_delay);
1006                         if (r < 0)
1007                                 log_warning("Invalid udev.exec-delay ignored: %s", value);
1008                 } else if ((value = startswith(opt, "udev.event-timeout="))) {
1009                         r = safe_atou64(value, &arg_event_timeout_usec);
1010                         if (r < 0) {
1011                                 log_warning("Invalid udev.event-timeout ignored: %s", value);
1012                                 break;
1013                         }
1014                         arg_event_timeout_usec *= USEC_PER_SEC;
1015                         arg_event_timeout_warn_usec = (arg_event_timeout_usec / 3) ? : 1;
1016                 }
1017
1018                 free(s);
1019         }
1020 }
1021
1022 static void help(void) {
1023         printf("%s [OPTIONS...]\n\n"
1024                "Manages devices.\n\n"
1025                "  -h --help                   Print this message\n"
1026                "     --version                Print version of the program\n"
1027                "     --daemon                 Detach and run in the background\n"
1028                "     --debug                  Enable debug output\n"
1029                "     --children-max=INT       Set maximum number of workers\n"
1030                "     --exec-delay=SECONDS     Seconds to wait before executing RUN=\n"
1031                "     --event-timeout=SECONDS  Seconds to wait before terminating an event\n"
1032                "     --resolve-names=early|late|never\n"
1033                "                              When to resolve users and groups\n"
1034                , program_invocation_short_name);
1035 }
1036
1037 static int parse_argv(int argc, char *argv[]) {
1038         static const struct option options[] = {
1039                 { "daemon",             no_argument,            NULL, 'd' },
1040                 { "debug",              no_argument,            NULL, 'D' },
1041                 { "children-max",       required_argument,      NULL, 'c' },
1042                 { "exec-delay",         required_argument,      NULL, 'e' },
1043                 { "event-timeout",      required_argument,      NULL, 't' },
1044                 { "resolve-names",      required_argument,      NULL, 'N' },
1045                 { "help",               no_argument,            NULL, 'h' },
1046                 { "version",            no_argument,            NULL, 'V' },
1047                 {}
1048         };
1049
1050         int c;
1051
1052         assert(argc >= 0);
1053         assert(argv);
1054
1055         while ((c = getopt_long(argc, argv, "c:de:DtN:hV", options, NULL)) >= 0) {
1056                 int r;
1057
1058                 switch (c) {
1059
1060                 case 'd':
1061                         arg_daemonize = true;
1062                         break;
1063                 case 'c':
1064                         r = safe_atoi(optarg, &arg_children_max);
1065                         if (r < 0)
1066                                 log_warning("Invalid --children-max ignored: %s", optarg);
1067                         break;
1068                 case 'e':
1069                         r = safe_atoi(optarg, &arg_exec_delay);
1070                         if (r < 0)
1071                                 log_warning("Invalid --exec-delay ignored: %s", optarg);
1072                         break;
1073                 case 't':
1074                         r = safe_atou64(optarg, &arg_event_timeout_usec);
1075                         if (r < 0)
1076                                 log_warning("Invalid --event-timeout ignored: %s", optarg);
1077                         else {
1078                                 arg_event_timeout_usec *= USEC_PER_SEC;
1079                                 arg_event_timeout_warn_usec = (arg_event_timeout_usec / 3) ? : 1;
1080                         }
1081                         break;
1082                 case 'D':
1083                         arg_debug = true;
1084                         break;
1085                 case 'N':
1086                         if (streq(optarg, "early")) {
1087                                 arg_resolve_names = 1;
1088                         } else if (streq(optarg, "late")) {
1089                                 arg_resolve_names = 0;
1090                         } else if (streq(optarg, "never")) {
1091                                 arg_resolve_names = -1;
1092                         } else {
1093                                 log_error("resolve-names must be early, late or never");
1094                                 return 0;
1095                         }
1096                         break;
1097                 case 'h':
1098                         help();
1099                         return 0;
1100                 case 'V':
1101                         printf("%s\n", VERSION);
1102                         return 0;
1103                 case '?':
1104                         return -EINVAL;
1105                 default:
1106                         assert_not_reached("Unhandled option");
1107
1108                 }
1109         }
1110
1111         return 1;
1112 }
1113
1114 int main(int argc, char *argv[]) {
1115         struct udev *udev;
1116         sigset_t mask;
1117         int fd_ctrl = -1;
1118         int fd_netlink = -1;
1119         int fd_worker = -1;
1120         struct epoll_event ep_ctrl = { .events = EPOLLIN };
1121         struct epoll_event ep_inotify = { .events = EPOLLIN };
1122         struct epoll_event ep_signal = { .events = EPOLLIN };
1123         struct epoll_event ep_netlink = { .events = EPOLLIN };
1124         struct epoll_event ep_worker = { .events = EPOLLIN };
1125         struct udev_ctrl_connection *ctrl_conn = NULL;
1126         int rc = 1, r;
1127
1128         udev = udev_new();
1129         if (udev == NULL)
1130                 goto exit;
1131
1132         log_set_target(LOG_TARGET_AUTO);
1133         log_parse_environment();
1134         log_open();
1135
1136         r = parse_argv(argc, argv);
1137         if (r <= 0)
1138                 goto exit;
1139
1140         kernel_cmdline_options(udev);
1141
1142         if (arg_debug)
1143                 log_set_max_level(LOG_DEBUG);
1144
1145         if (getuid() != 0) {
1146                 log_error("root privileges required");
1147                 goto exit;
1148         }
1149
1150         r = mac_selinux_init("/dev");
1151         if (r < 0) {
1152                 log_error_errno(r, "could not initialize labelling: %m");
1153                 goto exit;
1154         }
1155
1156         /* set umask before creating any file/directory */
1157         r = chdir("/");
1158         if (r < 0) {
1159                 log_error_errno(errno, "could not change dir to /: %m");
1160                 goto exit;
1161         }
1162
1163         umask(022);
1164
1165         udev_list_init(udev, &properties_list, true);
1166
1167         r = mkdir("/run/udev", 0755);
1168         if (r < 0 && errno != EEXIST) {
1169                 log_error_errno(errno, "could not create /run/udev: %m");
1170                 goto exit;
1171         }
1172
1173         dev_setup(NULL);
1174
1175         /* before opening new files, make sure std{in,out,err} fds are in a sane state */
1176         if (arg_daemonize) {
1177                 int fd;
1178
1179                 fd = open("/dev/null", O_RDWR);
1180                 if (fd >= 0) {
1181                         if (write(STDOUT_FILENO, 0, 0) < 0)
1182                                 dup2(fd, STDOUT_FILENO);
1183                         if (write(STDERR_FILENO, 0, 0) < 0)
1184                                 dup2(fd, STDERR_FILENO);
1185                         if (fd > STDERR_FILENO)
1186                                 close(fd);
1187                 } else {
1188                         log_error("cannot open /dev/null");
1189                 }
1190         }
1191
1192         if (systemd_fds(udev, &fd_ctrl, &fd_netlink) >= 0) {
1193                 /* get control and netlink socket from systemd */
1194                 udev_ctrl = udev_ctrl_new_from_fd(udev, fd_ctrl);
1195                 if (udev_ctrl == NULL) {
1196                         log_error("error taking over udev control socket");
1197                         rc = 1;
1198                         goto exit;
1199                 }
1200
1201                 monitor = udev_monitor_new_from_netlink_fd(udev, "kernel", fd_netlink);
1202                 if (monitor == NULL) {
1203                         log_error("error taking over netlink socket");
1204                         rc = 3;
1205                         goto exit;
1206                 }
1207
1208                 /* get our own cgroup, we regularly kill everything udev has left behind */
1209                 if (cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &udev_cgroup) < 0)
1210                         udev_cgroup = NULL;
1211         } else {
1212                 /* open control and netlink socket */
1213                 udev_ctrl = udev_ctrl_new(udev);
1214                 if (udev_ctrl == NULL) {
1215                         log_error("error initializing udev control socket");
1216                         rc = 1;
1217                         goto exit;
1218                 }
1219                 fd_ctrl = udev_ctrl_get_fd(udev_ctrl);
1220
1221                 monitor = udev_monitor_new_from_netlink(udev, "kernel");
1222                 if (monitor == NULL) {
1223                         log_error("error initializing netlink socket");
1224                         rc = 3;
1225                         goto exit;
1226                 }
1227                 fd_netlink = udev_monitor_get_fd(monitor);
1228
1229                 udev_monitor_set_receive_buffer_size(monitor, 128 * 1024 * 1024);
1230         }
1231
1232         if (udev_monitor_enable_receiving(monitor) < 0) {
1233                 log_error("error binding netlink socket");
1234                 rc = 3;
1235                 goto exit;
1236         }
1237
1238         if (udev_ctrl_enable_receiving(udev_ctrl) < 0) {
1239                 log_error("error binding udev control socket");
1240                 rc = 1;
1241                 goto exit;
1242         }
1243
1244         log_info("starting version " VERSION);
1245
1246         udev_builtin_init(udev);
1247
1248         rules = udev_rules_new(udev, arg_resolve_names);
1249         if (rules == NULL) {
1250                 log_error("error reading rules");
1251                 goto exit;
1252         }
1253
1254         rc = udev_rules_apply_static_dev_perms(rules);
1255         if (rc < 0)
1256                 log_error_errno(rc, "failed to apply permissions on static device nodes - %m");
1257
1258         if (arg_daemonize) {
1259                 pid_t pid;
1260
1261                 pid = fork();
1262                 switch (pid) {
1263                 case 0:
1264                         break;
1265                 case -1:
1266                         log_error_errno(errno, "fork of daemon failed: %m");
1267                         rc = 4;
1268                         goto exit;
1269                 default:
1270                         rc = EXIT_SUCCESS;
1271                         goto exit_daemonize;
1272                 }
1273
1274                 setsid();
1275
1276                 write_string_file("/proc/self/oom_score_adj", "-1000");
1277         } else {
1278                 sd_notify(1, "READY=1");
1279         }
1280
1281         if (arg_children_max <= 0) {
1282                 cpu_set_t cpu_set;
1283
1284                 arg_children_max = 8;
1285
1286                 if (sched_getaffinity(0, sizeof (cpu_set), &cpu_set) == 0) {
1287                         arg_children_max +=  CPU_COUNT(&cpu_set) * 2;
1288                 }
1289         }
1290         log_debug("set children_max to %u", arg_children_max);
1291
1292         udev_list_node_init(&event_list);
1293         udev_list_node_init(&worker_list);
1294
1295         fd_inotify = udev_watch_init(udev);
1296         if (fd_inotify < 0) {
1297                 log_error("error initializing inotify");
1298                 rc = 4;
1299                 goto exit;
1300         }
1301         udev_watch_restore(udev);
1302
1303         /* block and listen to all signals on signalfd */
1304         sigfillset(&mask);
1305         sigprocmask(SIG_SETMASK, &mask, &sigmask_orig);
1306         fd_signal = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
1307         if (fd_signal < 0) {
1308                 log_error("error creating signalfd");
1309                 rc = 5;
1310                 goto exit;
1311         }
1312
1313         /* unnamed socket from workers to the main daemon */
1314         if (socketpair(AF_LOCAL, SOCK_DGRAM|SOCK_CLOEXEC, 0, worker_watch) < 0) {
1315                 log_error("error creating socketpair");
1316                 rc = 6;
1317                 goto exit;
1318         }
1319         fd_worker = worker_watch[READ_END];
1320
1321         ep_ctrl.data.fd = fd_ctrl;
1322         ep_inotify.data.fd = fd_inotify;
1323         ep_signal.data.fd = fd_signal;
1324         ep_netlink.data.fd = fd_netlink;
1325         ep_worker.data.fd = fd_worker;
1326
1327         fd_ep = epoll_create1(EPOLL_CLOEXEC);
1328         if (fd_ep < 0) {
1329                 log_error_errno(errno, "error creating epoll fd: %m");
1330                 goto exit;
1331         }
1332         if (epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_ctrl, &ep_ctrl) < 0 ||
1333             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_inotify, &ep_inotify) < 0 ||
1334             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_signal, &ep_signal) < 0 ||
1335             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_netlink, &ep_netlink) < 0 ||
1336             epoll_ctl(fd_ep, EPOLL_CTL_ADD, fd_worker, &ep_worker) < 0) {
1337                 log_error_errno(errno, "fail to add fds to epoll: %m");
1338                 goto exit;
1339         }
1340
1341         for (;;) {
1342                 static usec_t last_usec;
1343                 struct epoll_event ev[8];
1344                 int fdcount;
1345                 int timeout;
1346                 bool is_worker, is_signal, is_inotify, is_netlink, is_ctrl;
1347                 int i;
1348
1349                 if (udev_exit) {
1350                         /* close sources of new events and discard buffered events */
1351                         if (fd_ctrl >= 0) {
1352                                 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_ctrl, NULL);
1353                                 fd_ctrl = -1;
1354                         }
1355                         if (monitor != NULL) {
1356                                 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_netlink, NULL);
1357                                 udev_monitor_unref(monitor);
1358                                 monitor = NULL;
1359                         }
1360                         if (fd_inotify >= 0) {
1361                                 epoll_ctl(fd_ep, EPOLL_CTL_DEL, fd_inotify, NULL);
1362                                 close(fd_inotify);
1363                                 fd_inotify = -1;
1364                         }
1365
1366                         /* discard queued events and kill workers */
1367                         event_queue_cleanup(udev, EVENT_QUEUED);
1368                         worker_kill(udev);
1369
1370                         /* exit after all has cleaned up */
1371                         if (udev_list_node_is_empty(&event_list) && children == 0)
1372                                 break;
1373
1374                         /* timeout at exit for workers to finish */
1375                         timeout = 30 * MSEC_PER_SEC;
1376                 } else if (udev_list_node_is_empty(&event_list) && children == 0) {
1377                         /* we are idle */
1378                         timeout = -1;
1379
1380                         /* cleanup possible left-over processes in our cgroup */
1381                         if (udev_cgroup)
1382                                 cg_kill(SYSTEMD_CGROUP_CONTROLLER, udev_cgroup, SIGKILL, false, true, NULL);
1383                 } else {
1384                         /* kill idle or hanging workers */
1385                         timeout = 3 * MSEC_PER_SEC;
1386                 }
1387
1388                 /* tell settle that we are busy or idle */
1389                 event_queue_update();
1390
1391                 fdcount = epoll_wait(fd_ep, ev, ELEMENTSOF(ev), timeout);
1392                 if (fdcount < 0)
1393                         continue;
1394
1395                 if (fdcount == 0) {
1396                         struct udev_list_node *loop;
1397
1398                         /* timeout */
1399                         if (udev_exit) {
1400                                 log_error("timeout, giving up waiting for workers to finish");
1401                                 break;
1402                         }
1403
1404                         /* kill idle workers */
1405                         if (udev_list_node_is_empty(&event_list)) {
1406                                 log_debug("cleanup idle workers");
1407                                 worker_kill(udev);
1408                         }
1409
1410                         /* check for hanging events */
1411                         udev_list_node_foreach(loop, &worker_list) {
1412                                 struct worker *worker = node_to_worker(loop);
1413                                 usec_t ts;
1414
1415                                 if (worker->state != WORKER_RUNNING)
1416                                         continue;
1417
1418                                 ts = now(CLOCK_MONOTONIC);
1419
1420                                 if ((ts - worker->event_start_usec) > arg_event_timeout_warn_usec) {
1421                                         if ((ts - worker->event_start_usec) > arg_event_timeout_usec) {
1422                                                 log_error("worker ["PID_FMT"] %s timeout; kill it", worker->pid, worker->event->devpath);
1423                                                 kill(worker->pid, SIGKILL);
1424                                                 worker->state = WORKER_KILLED;
1425
1426                                                 log_error("seq %llu '%s' killed", udev_device_get_seqnum(worker->event->dev), worker->event->devpath);
1427                                                 worker->event->exitcode = -64;
1428                                         } else if (!worker->event_warned) {
1429                                                 log_warning("worker ["PID_FMT"] %s is taking a long time", worker->pid, worker->event->devpath);
1430                                                 worker->event_warned = true;
1431                                         }
1432                                 }
1433                         }
1434
1435                 }
1436
1437                 is_worker = is_signal = is_inotify = is_netlink = is_ctrl = false;
1438                 for (i = 0; i < fdcount; i++) {
1439                         if (ev[i].data.fd == fd_worker && ev[i].events & EPOLLIN)
1440                                 is_worker = true;
1441                         else if (ev[i].data.fd == fd_netlink && ev[i].events & EPOLLIN)
1442                                 is_netlink = true;
1443                         else if (ev[i].data.fd == fd_signal && ev[i].events & EPOLLIN)
1444                                 is_signal = true;
1445                         else if (ev[i].data.fd == fd_inotify && ev[i].events & EPOLLIN)
1446                                 is_inotify = true;
1447                         else if (ev[i].data.fd == fd_ctrl && ev[i].events & EPOLLIN)
1448                                 is_ctrl = true;
1449                 }
1450
1451                 /* check for changed config, every 3 seconds at most */
1452                 if ((now(CLOCK_MONOTONIC) - last_usec) > 3 * USEC_PER_SEC) {
1453                         if (udev_rules_check_timestamp(rules))
1454                                 reload = true;
1455                         if (udev_builtin_validate(udev))
1456                                 reload = true;
1457
1458                         last_usec = now(CLOCK_MONOTONIC);
1459                 }
1460
1461                 /* reload requested, HUP signal received, rules changed, builtin changed */
1462                 if (reload) {
1463                         worker_kill(udev);
1464                         rules = udev_rules_unref(rules);
1465                         udev_builtin_exit(udev);
1466                         reload = false;
1467                 }
1468
1469                 /* event has finished */
1470                 if (is_worker)
1471                         worker_returned(fd_worker);
1472
1473                 if (is_netlink) {
1474                         struct udev_device *dev;
1475
1476                         dev = udev_monitor_receive_device(monitor);
1477                         if (dev != NULL) {
1478                                 udev_device_set_usec_initialized(dev, now(CLOCK_MONOTONIC));
1479                                 if (event_queue_insert(dev) < 0)
1480                                         udev_device_unref(dev);
1481                         }
1482                 }
1483
1484                 /* start new events */
1485                 if (!udev_list_node_is_empty(&event_list) && !udev_exit && !stop_exec_queue) {
1486                         udev_builtin_init(udev);
1487                         if (rules == NULL)
1488                                 rules = udev_rules_new(udev, arg_resolve_names);
1489                         if (rules != NULL)
1490                                 event_queue_start(udev);
1491                 }
1492
1493                 if (is_signal) {
1494                         struct signalfd_siginfo fdsi;
1495                         ssize_t size;
1496
1497                         size = read(fd_signal, &fdsi, sizeof(struct signalfd_siginfo));
1498                         if (size == sizeof(struct signalfd_siginfo))
1499                                 handle_signal(udev, fdsi.ssi_signo);
1500                 }
1501
1502                 /* we are shutting down, the events below are not handled anymore */
1503                 if (udev_exit)
1504                         continue;
1505
1506                 /* device node watch */
1507                 if (is_inotify)
1508                         handle_inotify(udev);
1509
1510                 /* tell settle that we are busy or idle, this needs to be before the
1511                  * PING handling
1512                  */
1513                 event_queue_update();
1514
1515                 /*
1516                  * This needs to be after the inotify handling, to make sure,
1517                  * that the ping is send back after the possibly generated
1518                  * "change" events by the inotify device node watch.
1519                  *
1520                  * A single time we may receive a client connection which we need to
1521                  * keep open to block the client. It will be closed right before we
1522                  * exit.
1523                  */
1524                 if (is_ctrl)
1525                         ctrl_conn = handle_ctrl_msg(udev_ctrl);
1526         }
1527
1528         rc = EXIT_SUCCESS;
1529 exit:
1530         udev_ctrl_cleanup(udev_ctrl);
1531         unlink("/run/udev/queue");
1532 exit_daemonize:
1533         if (fd_ep >= 0)
1534                 close(fd_ep);
1535         worker_list_cleanup(udev);
1536         event_queue_cleanup(udev, EVENT_UNDEF);
1537         udev_rules_unref(rules);
1538         udev_builtin_exit(udev);
1539         if (fd_signal >= 0)
1540                 close(fd_signal);
1541         if (worker_watch[READ_END] >= 0)
1542                 close(worker_watch[READ_END]);
1543         if (worker_watch[WRITE_END] >= 0)
1544                 close(worker_watch[WRITE_END]);
1545         udev_monitor_unref(monitor);
1546         udev_ctrl_connection_unref(ctrl_conn);
1547         udev_ctrl_unref(udev_ctrl);
1548         udev_list_cleanup(&properties_list);
1549         mac_selinux_finish();
1550         udev_unref(udev);
1551         log_close();
1552         return rc;
1553 }