src/udev/udevd.c

   1 /* SPDX-License-Identifier: GPL-2.0+ */
   2 /*
   3  * Copyright © 2004 Chris Friesen <chris_friesen@sympatico.ca>
   4  * Copyright © 2009 Canonical Ltd.
   5  * Copyright © 2009 Scott James Remnant <scott@netsplit.com>
   6  */
   7
   8 #include <errno.h>
   9 #include <fcntl.h>
  10 #include <getopt.h>
  11 #include <stdbool.h>
  12 #include <stddef.h>
  13 #include <stdio.h>
  14 #include <stdlib.h>
  15 #include <sys/epoll.h>
  16 #include <sys/file.h>
  17 #include <sys/inotify.h>
  18 #include <sys/ioctl.h>
  19 #include <sys/mount.h>
  20 #include <sys/prctl.h>
  21 #include <sys/signalfd.h>
  22 #include <sys/stat.h>
  23 #include <sys/time.h>
  24 #include <sys/wait.h>
  25 #include <unistd.h>
  26
  27 #include "sd-daemon.h"
  28 #include "sd-event.h"
  29
  30 #include "alloc-util.h"
  31 #include "build.h"
  32 #include "cgroup-util.h"
  33 #include "cpu-set-util.h"
  34 #include "dev-setup.h"
  35 #include "device-monitor-private.h"
  36 #include "device-private.h"
  37 #include "device-util.h"
  38 #include "event-util.h"
  39 #include "fd-util.h"
  40 #include "fileio.h"
  41 #include "format-util.h"
  42 #include "fs-util.h"
  43 #include "hashmap.h"
  44 #include "io-util.h"
  45 #include "libudev-device-internal.h"
  46 #include "limits-util.h"
  47 #include "list.h"
  48 #include "main-func.h"
  49 #include "mkdir.h"
  50 #include "netlink-util.h"
  51 #include "parse-util.h"
  52 #include "pretty-print.h"
  53 #include "proc-cmdline.h"
  54 #include "process-util.h"
  55 #include "selinux-util.h"
  56 #include "signal-util.h"
  57 #include "socket-util.h"
  58 #include "string-util.h"
  59 #include "strv.h"
  60 #include "strxcpyx.h"
  61 #include "syslog-util.h"
  62 #include "udevd.h"
  63 #include "udev-builtin.h"
  64 #include "udev-ctrl.h"
  65 #include "udev-event.h"
  66 #include "udev-util.h"
  67 #include "udev-watch.h"
  68 #include "user-util.h"
  69
  70 #define WORKER_NUM_MAX 2048U
  71
  72 static bool arg_debug = false;
  73 static int arg_daemonize = false;
  74 static ResolveNameTiming arg_resolve_name_timing = RESOLVE_NAME_EARLY;
  75 static unsigned arg_children_max = 0;
  76 static usec_t arg_exec_delay_usec = 0;
  77 static usec_t arg_event_timeout_usec = 180 * USEC_PER_SEC;
  78 static int arg_timeout_signal = SIGKILL;
  79 static bool arg_blockdev_read_only = false;
  80
  81 typedef struct Manager {
  82         sd_event *event;
  83         Hashmap *workers;
  84         LIST_HEAD(struct event, events);
  85         const char *cgroup;
  86         pid_t pid; /* the process that originally allocated the manager object */
  87
  88         UdevRules *rules;
  89         Hashmap *properties;
  90
  91         sd_netlink *rtnl;
  92
  93         sd_device_monitor *monitor;
  94         struct udev_ctrl *ctrl;
  95         int fd_inotify;
  96         int worker_watch[2];
  97
  98         sd_event_source *inotify_event;
  99         sd_event_source *kill_workers_event;
 100
 101         usec_t last_usec;
 102
 103         bool stop_exec_queue:1;
 104         bool exit:1;
 105 } Manager;
 106
 107 enum event_state {
 108         EVENT_UNDEF,
 109         EVENT_QUEUED,
 110         EVENT_RUNNING,
 111 };
 112
 113 struct event {
 114         Manager *manager;
 115         struct worker *worker;
 116         enum event_state state;
 117
 118         sd_device *dev;
 119         sd_device *dev_kernel; /* clone of originally received device */
 120
 121         uint64_t seqnum;
 122         uint64_t delaying_seqnum;
 123
 124         sd_event_source *timeout_warning_event;
 125         sd_event_source *timeout_event;
 126
 127         LIST_FIELDS(struct event, event);
 128 };
 129
 130 static void event_queue_cleanup(Manager *manager, enum event_state type);
 131
 132 enum worker_state {
 133         WORKER_UNDEF,
 134         WORKER_RUNNING,
 135         WORKER_IDLE,
 136         WORKER_KILLED,
 137 };
 138
 139 struct worker {
 140         Manager *manager;
 141         pid_t pid;
 142         sd_device_monitor *monitor;
 143         enum worker_state state;
 144         struct event *event;
 145 };
 146
 147 /* passed from worker to main process */
 148 struct worker_message {
 149 };
 150
 151 static void event_free(struct event *event) {
 152         if (!event)
 153                 return;
 154
 155         assert(event->manager);
 156
 157         LIST_REMOVE(event, event->manager->events, event);
 158         sd_device_unref(event->dev);
 159         sd_device_unref(event->dev_kernel);
 160
 161         sd_event_source_unref(event->timeout_warning_event);
 162         sd_event_source_unref(event->timeout_event);
 163
 164         if (event->worker)
 165                 event->worker->event = NULL;
 166
 167         /* only clean up the queue from the process that created it */
 168         if (LIST_IS_EMPTY(event->manager->events) &&
 169             event->manager->pid == getpid_cached())
 170                 if (unlink("/run/udev/queue") < 0)
 171                         log_warning_errno(errno, "Failed to unlink /run/udev/queue: %m");
 172
 173         free(event);
 174 }
 175
 176 static void worker_free(struct worker *worker) {
 177         if (!worker)
 178                 return;
 179
 180         assert(worker->manager);
 181
 182         hashmap_remove(worker->manager->workers, PID_TO_PTR(worker->pid));
 183         sd_device_monitor_unref(worker->monitor);
 184         event_free(worker->event);
 185
 186         free(worker);
 187 }
 188
 189 DEFINE_TRIVIAL_CLEANUP_FUNC(struct worker *, worker_free);
 190 DEFINE_PRIVATE_HASH_OPS_WITH_VALUE_DESTRUCTOR(worker_hash_op, void, trivial_hash_func, trivial_compare_func, struct worker, worker_free);
 191
 192 static int worker_new(struct worker **ret, Manager *manager, sd_device_monitor *worker_monitor, pid_t pid) {
 193         _cleanup_(worker_freep) struct worker *worker = NULL;
 194         int r;
 195
 196         assert(ret);
 197         assert(manager);
 198         assert(worker_monitor);
 199         assert(pid > 1);
 200
 201         /* close monitor, but keep address around */
 202         device_monitor_disconnect(worker_monitor);
 203
 204         worker = new(struct worker, 1);
 205         if (!worker)
 206                 return -ENOMEM;
 207
 208         *worker = (struct worker) {
 209                 .manager = manager,
 210                 .monitor = sd_device_monitor_ref(worker_monitor),
 211                 .pid = pid,
 212         };
 213
 214         r = hashmap_ensure_allocated(&manager->workers, &worker_hash_op);
 215         if (r < 0)
 216                 return r;
 217
 218         r = hashmap_put(manager->workers, PID_TO_PTR(pid), worker);
 219         if (r < 0)
 220                 return r;
 221
 222         *ret = TAKE_PTR(worker);
 223
 224         return 0;
 225 }
 226
 227 static int on_event_timeout(sd_event_source *s, uint64_t usec, void *userdata) {
 228         struct event *event = userdata;
 229
 230         assert(event);
 231         assert(event->worker);
 232
 233         kill_and_sigcont(event->worker->pid, arg_timeout_signal);
 234         event->worker->state = WORKER_KILLED;
 235
 236         log_device_error(event->dev, "Worker ["PID_FMT"] processing SEQNUM=%"PRIu64" killed", event->worker->pid, event->seqnum);
 237
 238         return 1;
 239 }
 240
 241 static int on_event_timeout_warning(sd_event_source *s, uint64_t usec, void *userdata) {
 242         struct event *event = userdata;
 243
 244         assert(event);
 245         assert(event->worker);
 246
 247         log_device_warning(event->dev, "Worker ["PID_FMT"] processing SEQNUM=%"PRIu64" is taking a long time", event->worker->pid, event->seqnum);
 248
 249         return 1;
 250 }
 251
 252 static void worker_attach_event(struct worker *worker, struct event *event) {
 253         sd_event *e;
 254
 255         assert(worker);
 256         assert(worker->manager);
 257         assert(event);
 258         assert(!event->worker);
 259         assert(!worker->event);
 260
 261         worker->state = WORKER_RUNNING;
 262         worker->event = event;
 263         event->state = EVENT_RUNNING;
 264         event->worker = worker;
 265
 266         e = worker->manager->event;
 267
 268         (void) sd_event_add_time_relative(e, &event->timeout_warning_event, CLOCK_MONOTONIC,
 269                                           udev_warn_timeout(arg_event_timeout_usec), USEC_PER_SEC,
 270                                           on_event_timeout_warning, event);
 271
 272         (void) sd_event_add_time_relative(e, &event->timeout_event, CLOCK_MONOTONIC,
 273                                           arg_event_timeout_usec, USEC_PER_SEC,
 274                                           on_event_timeout, event);
 275 }
 276
 277 static void manager_clear_for_worker(Manager *manager) {
 278         assert(manager);
 279
 280         manager->inotify_event = sd_event_source_unref(manager->inotify_event);
 281         manager->kill_workers_event = sd_event_source_unref(manager->kill_workers_event);
 282
 283         manager->event = sd_event_unref(manager->event);
 284
 285         manager->workers = hashmap_free(manager->workers);
 286         event_queue_cleanup(manager, EVENT_UNDEF);
 287
 288         manager->monitor = sd_device_monitor_unref(manager->monitor);
 289         manager->ctrl = udev_ctrl_unref(manager->ctrl);
 290
 291         manager->worker_watch[READ_END] = safe_close(manager->worker_watch[READ_END]);
 292 }
 293
 294 static void manager_free(Manager *manager) {
 295         if (!manager)
 296                 return;
 297
 298         udev_builtin_exit();
 299
 300         if (manager->pid == getpid_cached())
 301                 udev_ctrl_cleanup(manager->ctrl);
 302
 303         manager_clear_for_worker(manager);
 304
 305         sd_netlink_unref(manager->rtnl);
 306
 307         hashmap_free_free_free(manager->properties);
 308         udev_rules_free(manager->rules);
 309
 310         safe_close(manager->fd_inotify);
 311         safe_close_pair(manager->worker_watch);
 312
 313         free(manager);
 314 }
 315
 316 DEFINE_TRIVIAL_CLEANUP_FUNC(Manager*, manager_free);
 317
 318 static int worker_send_message(int fd) {
 319         struct worker_message message = {};
 320
 321         return loop_write(fd, &message, sizeof(message), false);
 322 }
 323
 324 static int worker_lock_block_device(sd_device *dev, int *ret_fd) {
 325         _cleanup_close_ int fd = -1;
 326         const char *val;
 327         int r;
 328
 329         assert(dev);
 330         assert(ret_fd);
 331
 332         /*
 333          * Take a shared lock on the device node; this establishes
 334          * a concept of device "ownership" to serialize device
 335          * access. External processes holding an exclusive lock will
 336          * cause udev to skip the event handling; in the case udev
 337          * acquired the lock, the external process can block until
 338          * udev has finished its event handling.
 339          */
 340
 341         if (device_for_action(dev, DEVICE_ACTION_REMOVE))
 342                 return 0;
 343
 344         r = sd_device_get_subsystem(dev, &val);
 345         if (r < 0)
 346                 return log_device_debug_errno(dev, r, "Failed to get subsystem: %m");
 347
 348         if (!streq(val, "block"))
 349                 return 0;
 350
 351         r = sd_device_get_sysname(dev, &val);
 352         if (r < 0)
 353                 return log_device_debug_errno(dev, r, "Failed to get sysname: %m");
 354
 355         if (STARTSWITH_SET(val, "dm-", "md", "drbd"))
 356                 return 0;
 357
 358         r = sd_device_get_devtype(dev, &val);
 359         if (r < 0 && r != -ENOENT)
 360                 return log_device_debug_errno(dev, r, "Failed to get devtype: %m");
 361         if (r >= 0 && streq(val, "partition")) {
 362                 r = sd_device_get_parent(dev, &dev);
 363                 if (r < 0)
 364                         return log_device_debug_errno(dev, r, "Failed to get parent device: %m");
 365         }
 366
 367         r = sd_device_get_devname(dev, &val);
 368         if (r == -ENOENT)
 369                 return 0;
 370         if (r < 0)
 371                 return log_device_debug_errno(dev, r, "Failed to get devname: %m");
 372
 373         fd = open(val, O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
 374         if (fd < 0) {
 375                 log_device_debug_errno(dev, errno, "Failed to open '%s', ignoring: %m", val);
 376                 return 0;
 377         }
 378
 379         if (flock(fd, LOCK_SH|LOCK_NB) < 0)
 380                 return log_device_debug_errno(dev, errno, "Failed to flock(%s): %m", val);
 381
 382         *ret_fd = TAKE_FD(fd);
 383         return 1;
 384 }
 385
 386 static int worker_mark_block_device_read_only(sd_device *dev) {
 387         _cleanup_close_ int fd = -1;
 388         const char *val;
 389         int state = 1, r;
 390
 391         assert(dev);
 392
 393         if (!arg_blockdev_read_only)
 394                 return 0;
 395
 396         /* Do this only once, when the block device is new. If the device is later retriggered let's not
 397          * toggle the bit again, so that people can boot up with full read-only mode and then unset the bit
 398          * for specific devices only. */
 399         if (!device_for_action(dev, DEVICE_ACTION_ADD))
 400                 return 0;
 401
 402         r = sd_device_get_subsystem(dev, &val);
 403         if (r < 0)
 404                 return log_device_debug_errno(dev, r, "Failed to get subsystem: %m");
 405
 406         if (!streq(val, "block"))
 407                 return 0;
 408
 409         r = sd_device_get_sysname(dev, &val);
 410         if (r < 0)
 411                 return log_device_debug_errno(dev, r, "Failed to get sysname: %m");
 412
 413         /* Exclude synthetic devices for now, this is supposed to be a safety feature to avoid modification
 414          * of physical devices, and what sits on top of those doesn't really matter if we don't allow the
 415          * underlying block devices to receive changes. */
 416         if (STARTSWITH_SET(val, "dm-", "md", "drbd", "loop", "nbd", "zram"))
 417                 return 0;
 418
 419         r = sd_device_get_devname(dev, &val);
 420         if (r == -ENOENT)
 421                 return 0;
 422         if (r < 0)
 423                 return log_device_debug_errno(dev, r, "Failed to get devname: %m");
 424
 425         fd = open(val, O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
 426         if (fd < 0)
 427                 return log_device_debug_errno(dev, errno, "Failed to open '%s', ignoring: %m", val);
 428
 429         if (ioctl(fd, BLKROSET, &state) < 0)
 430                 return log_device_warning_errno(dev, errno, "Failed to mark block device '%s' read-only: %m", val);
 431
 432         log_device_info(dev, "Successfully marked block device '%s' read-only.", val);
 433         return 0;
 434 }
 435
 436 static int worker_process_device(Manager *manager, sd_device *dev) {
 437         _cleanup_(udev_event_freep) UdevEvent *udev_event = NULL;
 438         _cleanup_close_ int fd_lock = -1;
 439         DeviceAction action;
 440         uint64_t seqnum;
 441         int r;
 442
 443         assert(manager);
 444         assert(dev);
 445
 446         r = device_get_seqnum(dev, &seqnum);
 447         if (r < 0)
 448                 return log_device_debug_errno(dev, r, "Failed to get SEQNUM: %m");
 449
 450         r = device_get_action(dev, &action);
 451         if (r < 0)
 452                 return log_device_debug_errno(dev, r, "Failed to get ACTION: %m");
 453
 454         log_device_debug(dev, "Processing device (SEQNUM=%"PRIu64", ACTION=%s)",
 455                          seqnum, device_action_to_string(action));
 456
 457         udev_event = udev_event_new(dev, arg_exec_delay_usec, manager->rtnl);
 458         if (!udev_event)
 459                 return -ENOMEM;
 460
 461         r = worker_lock_block_device(dev, &fd_lock);
 462         if (r < 0)
 463                 return r;
 464
 465         (void) worker_mark_block_device_read_only(dev);
 466
 467         /* apply rules, create node, symlinks */
 468         r = udev_event_execute_rules(udev_event, arg_event_timeout_usec, arg_timeout_signal, manager->properties, manager->rules);
 469         if (r < 0)
 470                 return r;
 471
 472         udev_event_execute_run(udev_event, arg_event_timeout_usec, arg_timeout_signal);
 473
 474         if (!manager->rtnl)
 475                 /* in case rtnl was initialized */
 476                 manager->rtnl = sd_netlink_ref(udev_event->rtnl);
 477
 478         /* apply/restore inotify watch */
 479         if (udev_event->inotify_watch) {
 480                 (void) udev_watch_begin(dev);
 481                 r = device_update_db(dev);
 482                 if (r < 0)
 483                         return log_device_debug_errno(dev, r, "Failed to update database under /run/udev/data/: %m");
 484         }
 485
 486         log_device_debug(dev, "Device (SEQNUM=%"PRIu64", ACTION=%s) processed",
 487                          seqnum, device_action_to_string(action));
 488
 489         return 0;
 490 }
 491
 492 static int worker_device_monitor_handler(sd_device_monitor *monitor, sd_device *dev, void *userdata) {
 493         Manager *manager = userdata;
 494         int r;
 495
 496         assert(dev);
 497         assert(manager);
 498
 499         r = worker_process_device(manager, dev);
 500         if (r == -EAGAIN)
 501                 /* if we couldn't acquire the flock(), then proceed quietly */
 502                 log_device_debug_errno(dev, r, "Device currently locked, not processing.");
 503         else {
 504                 if (r < 0)
 505                         log_device_warning_errno(dev, r, "Failed to process device, ignoring: %m");
 506
 507                 /* send processed event back to libudev listeners */
 508                 r = device_monitor_send_device(monitor, NULL, dev);
 509                 if (r < 0)
 510                         log_device_warning_errno(dev, r, "Failed to send device, ignoring: %m");
 511         }
 512
 513         /* send udevd the result of the event execution */
 514         r = worker_send_message(manager->worker_watch[WRITE_END]);
 515         if (r < 0)
 516                 log_device_warning_errno(dev, r, "Failed to send signal to main daemon, ignoring: %m");
 517
 518         return 1;
 519 }
 520
 521 static int worker_main(Manager *_manager, sd_device_monitor *monitor, sd_device *first_device) {
 522         _cleanup_(sd_device_unrefp) sd_device *dev = first_device;
 523         _cleanup_(manager_freep) Manager *manager = _manager;
 524         int r;
 525
 526         assert(manager);
 527         assert(monitor);
 528         assert(dev);
 529
 530         unsetenv("NOTIFY_SOCKET");
 531
 532         assert_se(sigprocmask_many(SIG_BLOCK, NULL, SIGTERM, -1) >= 0);
 533
 534         /* Reset OOM score, we only protect the main daemon. */
 535         r = set_oom_score_adjust(0);
 536         if (r < 0)
 537                 log_debug_errno(r, "Failed to reset OOM score, ignoring: %m");
 538
 539         /* Clear unnecessary data in Manager object.*/
 540         manager_clear_for_worker(manager);
 541
 542         r = sd_event_new(&manager->event);
 543         if (r < 0)
 544                 return log_error_errno(r, "Failed to allocate event loop: %m");
 545
 546         r = sd_event_add_signal(manager->event, NULL, SIGTERM, NULL, NULL);
 547         if (r < 0)
 548                 return log_error_errno(r, "Failed to set SIGTERM event: %m");
 549
 550         r = sd_device_monitor_attach_event(monitor, manager->event);
 551         if (r < 0)
 552                 return log_error_errno(r, "Failed to attach event loop to device monitor: %m");
 553
 554         r = sd_device_monitor_start(monitor, worker_device_monitor_handler, manager);
 555         if (r < 0)
 556                 return log_error_errno(r, "Failed to start device monitor: %m");
 557
 558         (void) sd_event_source_set_description(sd_device_monitor_get_event_source(monitor), "worker-device-monitor");
 559
 560         /* Process first device */
 561         (void) worker_device_monitor_handler(monitor, dev, manager);
 562
 563         r = sd_event_loop(manager->event);
 564         if (r < 0)
 565                 return log_error_errno(r, "Event loop failed: %m");
 566
 567         return 0;
 568 }
 569
 570 static int worker_spawn(Manager *manager, struct event *event) {
 571         _cleanup_(sd_device_monitor_unrefp) sd_device_monitor *worker_monitor = NULL;
 572         struct worker *worker;
 573         pid_t pid;
 574         int r;
 575
 576         /* listen for new events */
 577         r = device_monitor_new_full(&worker_monitor, MONITOR_GROUP_NONE, -1);
 578         if (r < 0)
 579                 return r;
 580
 581         /* allow the main daemon netlink address to send devices to the worker */
 582         r = device_monitor_allow_unicast_sender(worker_monitor, manager->monitor);
 583         if (r < 0)
 584                 return log_error_errno(r, "Worker: Failed to set unicast sender: %m");
 585
 586         r = device_monitor_enable_receiving(worker_monitor);
 587         if (r < 0)
 588                 return log_error_errno(r, "Worker: Failed to enable receiving of device: %m");
 589
 590         r = safe_fork(NULL, FORK_DEATHSIG, &pid);
 591         if (r < 0) {
 592                 event->state = EVENT_QUEUED;
 593                 return log_error_errno(r, "Failed to fork() worker: %m");
 594         }
 595         if (r == 0) {
 596                 /* Worker process */
 597                 r = worker_main(manager, worker_monitor, sd_device_ref(event->dev));
 598                 log_close();
 599                 _exit(r < 0 ? EXIT_FAILURE : EXIT_SUCCESS);
 600         }
 601
 602         r = worker_new(&worker, manager, worker_monitor, pid);
 603         if (r < 0)
 604                 return log_error_errno(r, "Failed to create worker object: %m");
 605
 606         worker_attach_event(worker, event);
 607
 608         log_device_debug(event->dev, "Worker ["PID_FMT"] is forked for processing SEQNUM=%"PRIu64".", pid, event->seqnum);
 609         return 0;
 610 }
 611
 612 static void event_run(Manager *manager, struct event *event) {
 613         static bool log_children_max_reached = true;
 614         struct worker *worker;
 615         Iterator i;
 616         int r;
 617
 618         assert(manager);
 619         assert(event);
 620
 621         if (DEBUG_LOGGING) {
 622                 DeviceAction action;
 623
 624                 r = device_get_action(event->dev, &action);
 625                 log_device_debug(event->dev, "Device (SEQNUM=%"PRIu64", ACTION=%s) ready for processing",
 626                                  event->seqnum, r >= 0 ? device_action_to_string(action) : "<unknown>");
 627         }
 628
 629         HASHMAP_FOREACH(worker, manager->workers, i) {
 630                 if (worker->state != WORKER_IDLE)
 631                         continue;
 632
 633                 r = device_monitor_send_device(manager->monitor, worker->monitor, event->dev);
 634                 if (r < 0) {
 635                         log_device_error_errno(event->dev, r, "Worker ["PID_FMT"] did not accept message, killing the worker: %m",
 636                                                worker->pid);
 637                         (void) kill(worker->pid, SIGKILL);
 638                         worker->state = WORKER_KILLED;
 639                         continue;
 640                 }
 641                 worker_attach_event(worker, event);
 642                 return;
 643         }
 644
 645         if (hashmap_size(manager->workers) >= arg_children_max) {
 646
 647                 /* Avoid spamming the debug logs if the limit is already reached and
 648                  * many events still need to be processed */
 649                 if (log_children_max_reached && arg_children_max > 1) {
 650                         log_debug("Maximum number (%u) of children reached.", hashmap_size(manager->workers));
 651                         log_children_max_reached = false;
 652                 }
 653                 return;
 654         }
 655
 656         /* Re-enable the debug message for the next batch of events */
 657         log_children_max_reached = true;
 658
 659         /* start new worker and pass initial device */
 660         worker_spawn(manager, event);
 661 }
 662
 663 static int event_queue_insert(Manager *manager, sd_device *dev) {
 664         _cleanup_(sd_device_unrefp) sd_device *clone = NULL;
 665         struct event *event;
 666         DeviceAction action;
 667         uint64_t seqnum;
 668         int r;
 669
 670         assert(manager);
 671         assert(dev);
 672
 673         /* only one process can add events to the queue */
 674         assert(manager->pid == getpid_cached());
 675
 676         /* We only accepts devices received by device monitor. */
 677         r = device_get_seqnum(dev, &seqnum);
 678         if (r < 0)
 679                 return r;
 680
 681         /* Refuse devices do not have ACTION property. */
 682         r = device_get_action(dev, &action);
 683         if (r < 0)
 684                 return r;
 685
 686         /* Save original device to restore the state on failures. */
 687         r = device_shallow_clone(dev, &clone);
 688         if (r < 0)
 689                 return r;
 690
 691         r = device_copy_properties(clone, dev);
 692         if (r < 0)
 693                 return r;
 694
 695         event = new(struct event, 1);
 696         if (!event)
 697                 return -ENOMEM;
 698
 699         *event = (struct event) {
 700                 .manager = manager,
 701                 .dev = sd_device_ref(dev),
 702                 .dev_kernel = TAKE_PTR(clone),
 703                 .seqnum = seqnum,
 704                 .state = EVENT_QUEUED,
 705         };
 706
 707         if (LIST_IS_EMPTY(manager->events)) {
 708                 r = touch("/run/udev/queue");
 709                 if (r < 0)
 710                         log_warning_errno(r, "Failed to touch /run/udev/queue: %m");
 711         }
 712
 713         LIST_APPEND(event, manager->events, event);
 714
 715         log_device_debug(dev, "Device (SEQNUM=%"PRIu64", ACTION=%s) is queued",
 716                          seqnum, device_action_to_string(action));
 717
 718         return 0;
 719 }
 720
 721 static void manager_kill_workers(Manager *manager) {
 722         struct worker *worker;
 723         Iterator i;
 724
 725         assert(manager);
 726
 727         HASHMAP_FOREACH(worker, manager->workers, i) {
 728                 if (worker->state == WORKER_KILLED)
 729                         continue;
 730
 731                 worker->state = WORKER_KILLED;
 732                 (void) kill(worker->pid, SIGTERM);
 733         }
 734 }
 735
 736 /* lookup event for identical, parent, child device */
 737 static int is_device_busy(Manager *manager, struct event *event) {
 738         const char *subsystem, *devpath, *devpath_old = NULL;
 739         dev_t devnum = makedev(0, 0);
 740         struct event *loop_event;
 741         size_t devpath_len;
 742         int r, ifindex = 0;
 743         bool is_block;
 744
 745         r = sd_device_get_subsystem(event->dev, &subsystem);
 746         if (r < 0)
 747                 return r;
 748
 749         is_block = streq(subsystem, "block");
 750
 751         r = sd_device_get_devpath(event->dev, &devpath);
 752         if (r < 0)
 753                 return r;
 754
 755         devpath_len = strlen(devpath);
 756
 757         r = sd_device_get_property_value(event->dev, "DEVPATH_OLD", &devpath_old);
 758         if (r < 0 && r != -ENOENT)
 759                 return r;
 760
 761         r = sd_device_get_devnum(event->dev, &devnum);
 762         if (r < 0 && r != -ENOENT)
 763                 return r;
 764
 765         r = sd_device_get_ifindex(event->dev, &ifindex);
 766         if (r < 0 && r != -ENOENT)
 767                 return r;
 768
 769         /* check if queue contains events we depend on */
 770         LIST_FOREACH(event, loop_event, manager->events) {
 771                 size_t loop_devpath_len, common;
 772                 const char *loop_devpath;
 773
 774                 /* we already found a later event, earlier cannot block us, no need to check again */
 775                 if (loop_event->seqnum < event->delaying_seqnum)
 776                         continue;
 777
 778                 /* event we checked earlier still exists, no need to check again */
 779                 if (loop_event->seqnum == event->delaying_seqnum)
 780                         return true;
 781
 782                 /* found ourself, no later event can block us */
 783                 if (loop_event->seqnum >= event->seqnum)
 784                         break;
 785
 786                 /* check major/minor */
 787                 if (major(devnum) != 0) {
 788                         const char *s;
 789                         dev_t d;
 790
 791                         if (sd_device_get_subsystem(loop_event->dev, &s) < 0)
 792                                 continue;
 793
 794                         if (sd_device_get_devnum(loop_event->dev, &d) >= 0 &&
 795                             devnum == d && is_block == streq(s, "block"))
 796                                 goto set_delaying_seqnum;
 797                 }
 798
 799                 /* check network device ifindex */
 800                 if (ifindex > 0) {
 801                         int i;
 802
 803                         if (sd_device_get_ifindex(loop_event->dev, &i) >= 0 &&
 804                             ifindex == i)
 805                                 goto set_delaying_seqnum;
 806                 }
 807
 808                 if (sd_device_get_devpath(loop_event->dev, &loop_devpath) < 0)
 809                         continue;
 810
 811                 /* check our old name */
 812                 if (devpath_old && streq(devpath_old, loop_devpath))
 813                         goto set_delaying_seqnum;
 814
 815                 loop_devpath_len = strlen(loop_devpath);
 816
 817                 /* compare devpath */
 818                 common = MIN(devpath_len, loop_devpath_len);
 819
 820                 /* one devpath is contained in the other? */
 821                 if (!strneq(devpath, loop_devpath, common))
 822                         continue;
 823
 824                 /* identical device event found */
 825                 if (devpath_len == loop_devpath_len)
 826                         goto set_delaying_seqnum;
 827
 828                 /* parent device event found */
 829                 if (devpath[common] == '/')
 830                         goto set_delaying_seqnum;
 831
 832                 /* child device event found */
 833                 if (loop_devpath[common] == '/')
 834                         goto set_delaying_seqnum;
 835         }
 836
 837         return false;
 838
 839 set_delaying_seqnum:
 840         log_device_debug(event->dev, "SEQNUM=%" PRIu64 " blocked by SEQNUM=%" PRIu64,
 841                          event->seqnum, loop_event->seqnum);
 842
 843         event->delaying_seqnum = loop_event->seqnum;
 844         return true;
 845 }
 846
 847 static void manager_exit(Manager *manager) {
 848         assert(manager);
 849
 850         manager->exit = true;
 851
 852         sd_notify(false,
 853                   "STOPPING=1\n"
 854                   "STATUS=Starting shutdown...");
 855
 856         /* close sources of new events and discard buffered events */
 857         manager->ctrl = udev_ctrl_unref(manager->ctrl);
 858
 859         manager->inotify_event = sd_event_source_unref(manager->inotify_event);
 860         manager->fd_inotify = safe_close(manager->fd_inotify);
 861
 862         manager->monitor = sd_device_monitor_unref(manager->monitor);
 863
 864         /* discard queued events and kill workers */
 865         event_queue_cleanup(manager, EVENT_QUEUED);
 866         manager_kill_workers(manager);
 867 }
 868
 869 /* reload requested, HUP signal received, rules changed, builtin changed */
 870 static void manager_reload(Manager *manager) {
 871
 872         assert(manager);
 873
 874         sd_notify(false,
 875                   "RELOADING=1\n"
 876                   "STATUS=Flushing configuration...");
 877
 878         manager_kill_workers(manager);
 879         manager->rules = udev_rules_free(manager->rules);
 880         udev_builtin_exit();
 881
 882         sd_notifyf(false,
 883                    "READY=1\n"
 884                    "STATUS=Processing with %u children at max", arg_children_max);
 885 }
 886
 887 static int on_kill_workers_event(sd_event_source *s, uint64_t usec, void *userdata) {
 888         Manager *manager = userdata;
 889
 890         assert(manager);
 891
 892         log_debug("Cleanup idle workers");
 893         manager_kill_workers(manager);
 894
 895         return 1;
 896 }
 897
 898 static void event_queue_start(Manager *manager) {
 899         struct event *event;
 900         usec_t usec;
 901         int r;
 902
 903         assert(manager);
 904
 905         if (LIST_IS_EMPTY(manager->events) ||
 906             manager->exit || manager->stop_exec_queue)
 907                 return;
 908
 909         assert_se(sd_event_now(manager->event, CLOCK_MONOTONIC, &usec) >= 0);
 910         /* check for changed config, every 3 seconds at most */
 911         if (manager->last_usec == 0 ||
 912             usec - manager->last_usec > 3 * USEC_PER_SEC) {
 913                 if (udev_rules_check_timestamp(manager->rules) ||
 914                     udev_builtin_validate())
 915                         manager_reload(manager);
 916
 917                 manager->last_usec = usec;
 918         }
 919
 920         r = event_source_disable(manager->kill_workers_event);
 921         if (r < 0)
 922                 log_warning_errno(r, "Failed to disable event source for cleaning up idle workers, ignoring: %m");
 923
 924         udev_builtin_init();
 925
 926         if (!manager->rules) {
 927                 r = udev_rules_load(&manager->rules, arg_resolve_name_timing);
 928                 if (r < 0) {
 929                         log_warning_errno(r, "Failed to read udev rules: %m");
 930                         return;
 931                 }
 932         }
 933
 934         LIST_FOREACH(event, event, manager->events) {
 935                 if (event->state != EVENT_QUEUED)
 936                         continue;
 937
 938                 /* do not start event if parent or child event is still running */
 939                 if (is_device_busy(manager, event) != 0)
 940                         continue;
 941
 942                 event_run(manager, event);
 943         }
 944 }
 945
 946 static void event_queue_cleanup(Manager *manager, enum event_state match_type) {
 947         struct event *event, *tmp;
 948
 949         LIST_FOREACH_SAFE(event, event, tmp, manager->events) {
 950                 if (match_type != EVENT_UNDEF && match_type != event->state)
 951                         continue;
 952
 953                 event_free(event);
 954         }
 955 }
 956
 957 static int on_worker(sd_event_source *s, int fd, uint32_t revents, void *userdata) {
 958         Manager *manager = userdata;
 959
 960         assert(manager);
 961
 962         for (;;) {
 963                 struct worker_message msg;
 964                 struct iovec iovec = {
 965                         .iov_base = &msg,
 966                         .iov_len = sizeof(msg),
 967                 };
 968                 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred))) control;
 969                 struct msghdr msghdr = {
 970                         .msg_iov = &iovec,
 971                         .msg_iovlen = 1,
 972                         .msg_control = &control,
 973                         .msg_controllen = sizeof(control),
 974                 };
 975                 ssize_t size;
 976                 struct ucred *ucred;
 977                 struct worker *worker;
 978
 979                 size = recvmsg_safe(fd, &msghdr, MSG_DONTWAIT);
 980                 if (size == -EINTR)
 981                         continue;
 982                 if (size == -EAGAIN)
 983                         /* nothing more to read */
 984                         break;
 985                 if (size < 0)
 986                         return log_error_errno(size, "Failed to receive message: %m");
 987
 988                 cmsg_close_all(&msghdr);
 989
 990                 if (size != sizeof(struct worker_message)) {
 991                         log_warning("Ignoring worker message with invalid size %zi bytes", size);
 992                         continue;
 993                 }
 994
 995                 ucred = CMSG_FIND_DATA(&msghdr, SOL_SOCKET, SCM_CREDENTIALS, struct ucred);
 996                 if (!ucred || ucred->pid <= 0) {
 997                         log_warning("Ignoring worker message without valid PID");
 998                         continue;
 999                 }
1000
1001                 /* lookup worker who sent the signal */
1002                 worker = hashmap_get(manager->workers, PID_TO_PTR(ucred->pid));
1003                 if (!worker) {
1004                         log_debug("Worker ["PID_FMT"] returned, but is no longer tracked", ucred->pid);
1005                         continue;
1006                 }
1007
1008                 if (worker->state != WORKER_KILLED)
1009                         worker->state = WORKER_IDLE;
1010
1011                 /* worker returned */
1012                 event_free(worker->event);
1013         }
1014
1015         /* we have free workers, try to schedule events */
1016         event_queue_start(manager);
1017
1018         return 1;
1019 }
1020
1021 static int on_uevent(sd_device_monitor *monitor, sd_device *dev, void *userdata) {
1022         Manager *manager = userdata;
1023         int r;
1024
1025         assert(manager);
1026
1027         device_ensure_usec_initialized(dev, NULL);
1028
1029         r = event_queue_insert(manager, dev);
1030         if (r < 0) {
1031                 log_device_error_errno(dev, r, "Failed to insert device into event queue: %m");
1032                 return 1;
1033         }
1034
1035         /* we have fresh events, try to schedule them */
1036         event_queue_start(manager);
1037
1038         return 1;
1039 }
1040
1041 /* receive the udevd message from userspace */
1042 static int on_ctrl_msg(struct udev_ctrl *uctrl, enum udev_ctrl_msg_type type, const union udev_ctrl_msg_value *value, void *userdata) {
1043         Manager *manager = userdata;
1044         int r;
1045
1046         assert(value);
1047         assert(manager);
1048
1049         switch (type) {
1050         case UDEV_CTRL_SET_LOG_LEVEL:
1051                 log_debug("Received udev control message (SET_LOG_LEVEL), setting log_priority=%i", value->intval);
1052                 log_set_max_level_realm(LOG_REALM_UDEV, value->intval);
1053                 log_set_max_level_realm(LOG_REALM_SYSTEMD, value->intval);
1054                 manager_kill_workers(manager);
1055                 break;
1056         case UDEV_CTRL_STOP_EXEC_QUEUE:
1057                 log_debug("Received udev control message (STOP_EXEC_QUEUE)");
1058                 manager->stop_exec_queue = true;
1059                 break;
1060         case UDEV_CTRL_START_EXEC_QUEUE:
1061                 log_debug("Received udev control message (START_EXEC_QUEUE)");
1062                 manager->stop_exec_queue = false;
1063                 event_queue_start(manager);
1064                 break;
1065         case UDEV_CTRL_RELOAD:
1066                 log_debug("Received udev control message (RELOAD)");
1067                 manager_reload(manager);
1068                 break;
1069         case UDEV_CTRL_SET_ENV: {
1070                 _cleanup_free_ char *key = NULL, *val = NULL, *old_key = NULL, *old_val = NULL;
1071                 const char *eq;
1072
1073                 eq = strchr(value->buf, '=');
1074                 if (!eq) {
1075                         log_error("Invalid key format '%s'", value->buf);
1076                         return 1;
1077                 }
1078
1079                 key = strndup(value->buf, eq - value->buf);
1080                 if (!key) {
1081                         log_oom();
1082                         return 1;
1083                 }
1084
1085                 old_val = hashmap_remove2(manager->properties, key, (void **) &old_key);
1086
1087                 r = hashmap_ensure_allocated(&manager->properties, &string_hash_ops);
1088                 if (r < 0) {
1089                         log_oom();
1090                         return 1;
1091                 }
1092
1093                 eq++;
1094                 if (isempty(eq)) {
1095                         log_debug("Received udev control message (ENV), unsetting '%s'", key);
1096
1097                         r = hashmap_put(manager->properties, key, NULL);
1098                         if (r < 0) {
1099                                 log_oom();
1100                                 return 1;
1101                         }
1102                 } else {
1103                         val = strdup(eq);
1104                         if (!val) {
1105                                 log_oom();
1106                                 return 1;
1107                         }
1108
1109                         log_debug("Received udev control message (ENV), setting '%s=%s'", key, val);
1110
1111                         r = hashmap_put(manager->properties, key, val);
1112                         if (r < 0) {
1113                                 log_oom();
1114                                 return 1;
1115                         }
1116                 }
1117
1118                 key = val = NULL;
1119                 manager_kill_workers(manager);
1120                 break;
1121         }
1122         case UDEV_CTRL_SET_CHILDREN_MAX:
1123                 if (value->intval <= 0) {
1124                         log_debug("Received invalid udev control message (SET_MAX_CHILDREN, %i), ignoring.", value->intval);
1125                         return 0;
1126                 }
1127
1128                 log_debug("Received udev control message (SET_MAX_CHILDREN), setting children_max=%i", value->intval);
1129                 arg_children_max = value->intval;
1130
1131                 (void) sd_notifyf(false,
1132                                   "READY=1\n"
1133                                   "STATUS=Processing with %u children at max", arg_children_max);
1134                 break;
1135         case UDEV_CTRL_PING:
1136                 log_debug("Received udev control message (PING)");
1137                 break;
1138         case UDEV_CTRL_EXIT:
1139                 log_debug("Received udev control message (EXIT)");
1140                 manager_exit(manager);
1141                 break;
1142         default:
1143                 log_debug("Received unknown udev control message, ignoring");
1144         }
1145
1146         return 1;
1147 }
1148
1149 static int synthesize_change_one(sd_device *dev, const char *syspath) {
1150         const char *filename;
1151         int r;
1152
1153         filename = strjoina(syspath, "/uevent");
1154         log_device_debug(dev, "device is closed, synthesising 'change' on %s", syspath);
1155         r = write_string_file(filename, "change", WRITE_STRING_FILE_DISABLE_BUFFER);
1156         if (r < 0)
1157                 return log_device_debug_errno(dev, r, "Failed to write 'change' to %s: %m", filename);
1158         return 0;
1159 }
1160
1161 static int synthesize_change(sd_device *dev) {
1162         const char *subsystem, *sysname, *devname, *syspath, *devtype;
1163         int r;
1164
1165         r = sd_device_get_subsystem(dev, &subsystem);
1166         if (r < 0)
1167                 return r;
1168
1169         r = sd_device_get_sysname(dev, &sysname);
1170         if (r < 0)
1171                 return r;
1172
1173         r = sd_device_get_devname(dev, &devname);
1174         if (r < 0)
1175                 return r;
1176
1177         r = sd_device_get_syspath(dev, &syspath);
1178         if (r < 0)
1179                 return r;
1180
1181         r = sd_device_get_devtype(dev, &devtype);
1182         if (r < 0)
1183                 return r;
1184
1185         if (streq_ptr("block", subsystem) &&
1186             streq_ptr("disk", devtype) &&
1187             !startswith(sysname, "dm-")) {
1188                 _cleanup_(sd_device_enumerator_unrefp) sd_device_enumerator *e = NULL;
1189                 bool part_table_read = false, has_partitions = false;
1190                 sd_device *d;
1191                 int fd;
1192
1193                 /*
1194                  * Try to re-read the partition table. This only succeeds if
1195                  * none of the devices is busy. The kernel returns 0 if no
1196                  * partition table is found, and we will not get an event for
1197                  * the disk.
1198                  */
1199                 fd = open(devname, O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
1200                 if (fd >= 0) {
1201                         r = flock(fd, LOCK_EX|LOCK_NB);
1202                         if (r >= 0)
1203                                 r = ioctl(fd, BLKRRPART, 0);
1204
1205                         close(fd);
1206                         if (r >= 0)
1207                                 part_table_read = true;
1208                 }
1209
1210                 /* search for partitions */
1211                 r = sd_device_enumerator_new(&e);
1212                 if (r < 0)
1213                         return r;
1214
1215                 r = sd_device_enumerator_allow_uninitialized(e);
1216                 if (r < 0)
1217                         return r;
1218
1219                 r = sd_device_enumerator_add_match_parent(e, dev);
1220                 if (r < 0)
1221                         return r;
1222
1223                 r = sd_device_enumerator_add_match_subsystem(e, "block", true);
1224                 if (r < 0)
1225                         return r;
1226
1227                 FOREACH_DEVICE(e, d) {
1228                         const char *t;
1229
1230                         if (sd_device_get_devtype(d, &t) < 0 ||
1231                             !streq("partition", t))
1232                                 continue;
1233
1234                         has_partitions = true;
1235                         break;
1236                 }
1237
1238                 /*
1239                  * We have partitions and re-read the table, the kernel already sent
1240                  * out a "change" event for the disk, and "remove/add" for all
1241                  * partitions.
1242                  */
1243                 if (part_table_read && has_partitions)
1244                         return 0;
1245
1246                 /*
1247                  * We have partitions but re-reading the partition table did not
1248                  * work, synthesize "change" for the disk and all partitions.
1249                  */
1250                 (void) synthesize_change_one(dev, syspath);
1251
1252                 FOREACH_DEVICE(e, d) {
1253                         const char *t, *n, *s;
1254
1255                         if (sd_device_get_devtype(d, &t) < 0 ||
1256                             !streq("partition", t))
1257                                 continue;
1258
1259                         if (sd_device_get_devname(d, &n) < 0 ||
1260                             sd_device_get_syspath(d, &s) < 0)
1261                                 continue;
1262
1263                         (void) synthesize_change_one(dev, s);
1264                 }
1265
1266         } else
1267                 (void) synthesize_change_one(dev, syspath);
1268
1269         return 0;
1270 }
1271
1272 static int on_inotify(sd_event_source *s, int fd, uint32_t revents, void *userdata) {
1273         Manager *manager = userdata;
1274         union inotify_event_buffer buffer;
1275         struct inotify_event *e;
1276         ssize_t l;
1277         int r;
1278
1279         assert(manager);
1280
1281         r = event_source_disable(manager->kill_workers_event);
1282         if (r < 0)
1283                 log_warning_errno(r, "Failed to disable event source for cleaning up idle workers, ignoring: %m");
1284
1285         l = read(fd, &buffer, sizeof(buffer));
1286         if (l < 0) {
1287                 if (IN_SET(errno, EAGAIN, EINTR))
1288                         return 1;
1289
1290                 return log_error_errno(errno, "Failed to read inotify fd: %m");
1291         }
1292
1293         FOREACH_INOTIFY_EVENT(e, buffer, l) {
1294                 _cleanup_(sd_device_unrefp) sd_device *dev = NULL;
1295                 const char *devnode;
1296
1297                 if (udev_watch_lookup(e->wd, &dev) <= 0)
1298                         continue;
1299
1300                 if (sd_device_get_devname(dev, &devnode) < 0)
1301                         continue;
1302
1303                 log_device_debug(dev, "Inotify event: %x for %s", e->mask, devnode);
1304                 if (e->mask & IN_CLOSE_WRITE)
1305                         synthesize_change(dev);
1306                 else if (e->mask & IN_IGNORED)
1307                         udev_watch_end(dev);
1308         }
1309
1310         return 1;
1311 }
1312
1313 static int on_sigterm(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
1314         Manager *manager = userdata;
1315
1316         assert(manager);
1317
1318         manager_exit(manager);
1319
1320         return 1;
1321 }
1322
1323 static int on_sighup(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
1324         Manager *manager = userdata;
1325
1326         assert(manager);
1327
1328         manager_reload(manager);
1329
1330         return 1;
1331 }
1332
1333 static int on_sigchld(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
1334         Manager *manager = userdata;
1335         int r;
1336
1337         assert(manager);
1338
1339         for (;;) {
1340                 pid_t pid;
1341                 int status;
1342                 struct worker *worker;
1343
1344                 pid = waitpid(-1, &status, WNOHANG);
1345                 if (pid <= 0)
1346                         break;
1347
1348                 worker = hashmap_get(manager->workers, PID_TO_PTR(pid));
1349                 if (!worker) {
1350                         log_warning("Worker ["PID_FMT"] is unknown, ignoring", pid);
1351                         continue;
1352                 }
1353
1354                 if (WIFEXITED(status)) {
1355                         if (WEXITSTATUS(status) == 0)
1356                                 log_debug("Worker ["PID_FMT"] exited", pid);
1357                         else
1358                                 log_warning("Worker ["PID_FMT"] exited with return code %i", pid, WEXITSTATUS(status));
1359                 } else if (WIFSIGNALED(status))
1360                         log_warning("Worker ["PID_FMT"] terminated by signal %i (%s)", pid, WTERMSIG(status), signal_to_string(WTERMSIG(status)));
1361                 else if (WIFSTOPPED(status)) {
1362                         log_info("Worker ["PID_FMT"] stopped", pid);
1363                         continue;
1364                 } else if (WIFCONTINUED(status)) {
1365                         log_info("Worker ["PID_FMT"] continued", pid);
1366                         continue;
1367                 } else
1368                         log_warning("Worker ["PID_FMT"] exit with status 0x%04x", pid, status);
1369
1370                 if ((!WIFEXITED(status) || WEXITSTATUS(status) != 0) && worker->event) {
1371                         log_device_error(worker->event->dev, "Worker ["PID_FMT"] failed", pid);
1372
1373                         /* delete state from disk */
1374                         device_delete_db(worker->event->dev);
1375                         device_tag_index(worker->event->dev, NULL, false);
1376
1377                         if (manager->monitor) {
1378                                 /* forward kernel event without amending it */
1379                                 r = device_monitor_send_device(manager->monitor, NULL, worker->event->dev_kernel);
1380                                 if (r < 0)
1381                                         log_device_error_errno(worker->event->dev_kernel, r, "Failed to send back device to kernel: %m");
1382                         }
1383                 }
1384
1385                 worker_free(worker);
1386         }
1387
1388         /* we can start new workers, try to schedule events */
1389         event_queue_start(manager);
1390
1391         /* Disable unnecessary cleanup event */
1392         if (hashmap_isempty(manager->workers)) {
1393                 r = event_source_disable(manager->kill_workers_event);
1394                 if (r < 0)
1395                         log_warning_errno(r, "Failed to disable event source for cleaning up idle workers, ignoring: %m");
1396         }
1397
1398         return 1;
1399 }
1400
1401 static int on_post(sd_event_source *s, void *userdata) {
1402         Manager *manager = userdata;
1403
1404         assert(manager);
1405
1406         if (!LIST_IS_EMPTY(manager->events))
1407                 return 1;
1408
1409         /* There are no pending events. Let's cleanup idle process. */
1410
1411         if (!hashmap_isempty(manager->workers)) {
1412                 /* There are idle workers */
1413                 (void) event_reset_time(manager->event, &manager->kill_workers_event, CLOCK_MONOTONIC,
1414                                         now(CLOCK_MONOTONIC) + 3 * USEC_PER_SEC, USEC_PER_SEC,
1415                                         on_kill_workers_event, manager, 0, "kill-workers-event", false);
1416                 return 1;
1417         }
1418
1419         /* There are no idle workers. */
1420
1421         if (manager->exit)
1422                 return sd_event_exit(manager->event, 0);
1423
1424         if (manager->cgroup)
1425                 /* cleanup possible left-over processes in our cgroup */
1426                 (void) cg_kill(SYSTEMD_CGROUP_CONTROLLER, manager->cgroup, SIGKILL, CGROUP_IGNORE_SELF, NULL, NULL, NULL);
1427
1428         return 1;
1429 }
1430
1431 static int listen_fds(int *ret_ctrl, int *ret_netlink) {
1432         int ctrl_fd = -1, netlink_fd = -1;
1433         int fd, n;
1434
1435         assert(ret_ctrl);
1436         assert(ret_netlink);
1437
1438         n = sd_listen_fds(true);
1439         if (n < 0)
1440                 return n;
1441
1442         for (fd = SD_LISTEN_FDS_START; fd < n + SD_LISTEN_FDS_START; fd++) {
1443                 if (sd_is_socket(fd, AF_LOCAL, SOCK_SEQPACKET, -1) > 0) {
1444                         if (ctrl_fd >= 0)
1445                                 return -EINVAL;
1446                         ctrl_fd = fd;
1447                         continue;
1448                 }
1449
1450                 if (sd_is_socket(fd, AF_NETLINK, SOCK_RAW, -1) > 0) {
1451                         if (netlink_fd >= 0)
1452                                 return -EINVAL;
1453                         netlink_fd = fd;
1454                         continue;
1455                 }
1456
1457                 return -EINVAL;
1458         }
1459
1460         *ret_ctrl = ctrl_fd;
1461         *ret_netlink = netlink_fd;
1462
1463         return 0;
1464 }
1465
1466 /*
1467  * read the kernel command line, in case we need to get into debug mode
1468  *   udev.log_priority=<level>                 syslog priority
1469  *   udev.children_max=<number of workers>     events are fully serialized if set to 1
1470  *   udev.exec_delay=<number of seconds>       delay execution of every executed program
1471  *   udev.event_timeout=<number of seconds>    seconds to wait before terminating an event
1472  *   udev.blockdev_read_only<=bool>            mark all block devices read-only when they appear
1473  */
1474 static int parse_proc_cmdline_item(const char *key, const char *value, void *data) {
1475         int r;
1476
1477         assert(key);
1478
1479         if (proc_cmdline_key_streq(key, "udev.log_priority")) {
1480
1481                 if (proc_cmdline_value_missing(key, value))
1482                         return 0;
1483
1484                 r = log_level_from_string(value);
1485                 if (r >= 0)
1486                         log_set_max_level(r);
1487
1488         } else if (proc_cmdline_key_streq(key, "udev.event_timeout")) {
1489
1490                 if (proc_cmdline_value_missing(key, value))
1491                         return 0;
1492
1493                 r = parse_sec(value, &arg_event_timeout_usec);
1494
1495         } else if (proc_cmdline_key_streq(key, "udev.children_max")) {
1496
1497                 if (proc_cmdline_value_missing(key, value))
1498                         return 0;
1499
1500                 r = safe_atou(value, &arg_children_max);
1501
1502         } else if (proc_cmdline_key_streq(key, "udev.exec_delay")) {
1503
1504                 if (proc_cmdline_value_missing(key, value))
1505                         return 0;
1506
1507                 r = parse_sec(value, &arg_exec_delay_usec);
1508
1509         } else if (proc_cmdline_key_streq(key, "udev.timeout_signal")) {
1510
1511                 if (proc_cmdline_value_missing(key, value))
1512                         return 0;
1513
1514                 r = signal_from_string(value);
1515                 if (r > 0)
1516                         arg_timeout_signal = r;
1517
1518         } else if (proc_cmdline_key_streq(key, "udev.blockdev_read_only")) {
1519
1520                 if (!value)
1521                         arg_blockdev_read_only = true;
1522                 else {
1523                         r = parse_boolean(value);
1524                         if (r < 0)
1525                                 log_warning_errno(r, "Failed to parse udev.blockdev-read-only argument, ignoring: %s", value);
1526                         else
1527                                 arg_blockdev_read_only = r;
1528                 }
1529
1530                 if (arg_blockdev_read_only)
1531                         log_notice("All physical block devices will be marked read-only.");
1532
1533                 return 0;
1534
1535         } else {
1536                 if (startswith(key, "udev."))
1537                         log_warning("Unknown udev kernel command line option \"%s\", ignoring.", key);
1538
1539                 return 0;
1540         }
1541
1542         if (r < 0)
1543                 log_warning_errno(r, "Failed to parse \"%s=%s\", ignoring: %m", key, value);
1544
1545         return 0;
1546 }
1547
1548 static int help(void) {
1549         _cleanup_free_ char *link = NULL;
1550         int r;
1551
1552         r = terminal_urlify_man("systemd-udevd.service", "8", &link);
1553         if (r < 0)
1554                 return log_oom();
1555
1556         printf("%s [OPTIONS...]\n\n"
1557                "Rule-based manager for device events and files.\n\n"
1558                "  -h --help                   Print this message\n"
1559                "  -V --version                Print version of the program\n"
1560                "  -d --daemon                 Detach and run in the background\n"
1561                "  -D --debug                  Enable debug output\n"
1562                "  -c --children-max=INT       Set maximum number of workers\n"
1563                "  -e --exec-delay=SECONDS     Seconds to wait before executing RUN=\n"
1564                "  -t --event-timeout=SECONDS  Seconds to wait before terminating an event\n"
1565                "  -N --resolve-names=early|late|never\n"
1566                "                              When to resolve users and groups\n"
1567                "\nSee the %s for details.\n"
1568                , program_invocation_short_name
1569                , link
1570         );
1571
1572         return 0;
1573 }
1574
1575 static int parse_argv(int argc, char *argv[]) {
1576         enum {
1577                 ARG_TIMEOUT_SIGNAL,
1578         };
1579
1580         static const struct option options[] = {
1581                 { "daemon",             no_argument,            NULL, 'd'                 },
1582                 { "debug",              no_argument,            NULL, 'D'                 },
1583                 { "children-max",       required_argument,      NULL, 'c'                 },
1584                 { "exec-delay",         required_argument,      NULL, 'e'                 },
1585                 { "event-timeout",      required_argument,      NULL, 't'                 },
1586                 { "resolve-names",      required_argument,      NULL, 'N'                 },
1587                 { "help",               no_argument,            NULL, 'h'                 },
1588                 { "version",            no_argument,            NULL, 'V'                 },
1589                 { "timeout-signal",     required_argument,      NULL,  ARG_TIMEOUT_SIGNAL },
1590                 {}
1591         };
1592
1593         int c, r;
1594
1595         assert(argc >= 0);
1596         assert(argv);
1597
1598         while ((c = getopt_long(argc, argv, "c:de:Dt:N:hV", options, NULL)) >= 0) {
1599                 switch (c) {
1600
1601                 case 'd':
1602                         arg_daemonize = true;
1603                         break;
1604                 case 'c':
1605                         r = safe_atou(optarg, &arg_children_max);
1606                         if (r < 0)
1607                                 log_warning_errno(r, "Failed to parse --children-max= value '%s', ignoring: %m", optarg);
1608                         break;
1609                 case 'e':
1610                         r = parse_sec(optarg, &arg_exec_delay_usec);
1611                         if (r < 0)
1612                                 log_warning_errno(r, "Failed to parse --exec-delay= value '%s', ignoring: %m", optarg);
1613                         break;
1614                 case ARG_TIMEOUT_SIGNAL:
1615                         r = signal_from_string(optarg);
1616                         if (r <= 0)
1617                                 log_warning_errno(r, "Failed to parse --timeout-signal= value '%s', ignoring: %m", optarg);
1618                         else
1619                                 arg_timeout_signal = r;
1620
1621                         break;
1622                 case 't':
1623                         r = parse_sec(optarg, &arg_event_timeout_usec);
1624                         if (r < 0)
1625                                 log_warning_errno(r, "Failed to parse --event-timeout= value '%s', ignoring: %m", optarg);
1626                         break;
1627                 case 'D':
1628                         arg_debug = true;
1629                         break;
1630                 case 'N': {
1631                         ResolveNameTiming t;
1632
1633                         t = resolve_name_timing_from_string(optarg);
1634                         if (t < 0)
1635                                 log_warning("Invalid --resolve-names= value '%s', ignoring.", optarg);
1636                         else
1637                                 arg_resolve_name_timing = t;
1638                         break;
1639                 }
1640                 case 'h':
1641                         return help();
1642                 case 'V':
1643                         printf("%s\n", GIT_VERSION);
1644                         return 0;
1645                 case '?':
1646                         return -EINVAL;
1647                 default:
1648                         assert_not_reached("Unhandled option");
1649
1650                 }
1651         }
1652
1653         return 1;
1654 }
1655
1656 static int manager_new(Manager **ret, int fd_ctrl, int fd_uevent, const char *cgroup) {
1657         _cleanup_(manager_freep) Manager *manager = NULL;
1658         int r;
1659
1660         assert(ret);
1661
1662         manager = new(Manager, 1);
1663         if (!manager)
1664                 return log_oom();
1665
1666         *manager = (Manager) {
1667                 .fd_inotify = -1,
1668                 .worker_watch = { -1, -1 },
1669                 .cgroup = cgroup,
1670         };
1671
1672         r = udev_ctrl_new_from_fd(&manager->ctrl, fd_ctrl);
1673         if (r < 0)
1674                 return log_error_errno(r, "Failed to initialize udev control socket: %m");
1675
1676         r = udev_ctrl_enable_receiving(manager->ctrl);
1677         if (r < 0)
1678                 return log_error_errno(r, "Failed to bind udev control socket: %m");
1679
1680         r = device_monitor_new_full(&manager->monitor, MONITOR_GROUP_KERNEL, fd_uevent);
1681         if (r < 0)
1682                 return log_error_errno(r, "Failed to initialize device monitor: %m");
1683
1684         /* Bump receiver buffer, but only if we are not called via socket activation, as in that
1685          * case systemd sets the receive buffer size for us, and the value in the .socket unit
1686          * should take full effect. */
1687         if (fd_uevent < 0)
1688                 (void) sd_device_monitor_set_receive_buffer_size(manager->monitor, 128 * 1024 * 1024);
1689
1690         r = device_monitor_enable_receiving(manager->monitor);
1691         if (r < 0)
1692                 return log_error_errno(r, "Failed to bind netlink socket: %m");
1693
1694         *ret = TAKE_PTR(manager);
1695
1696         return 0;
1697 }
1698
1699 static int main_loop(Manager *manager) {
1700         int fd_worker, r;
1701
1702         manager->pid = getpid_cached();
1703
1704         /* unnamed socket from workers to the main daemon */
1705         r = socketpair(AF_LOCAL, SOCK_DGRAM|SOCK_CLOEXEC, 0, manager->worker_watch);
1706         if (r < 0)
1707                 return log_error_errno(errno, "Failed to create socketpair for communicating with workers: %m");
1708
1709         fd_worker = manager->worker_watch[READ_END];
1710
1711         r = setsockopt_int(fd_worker, SOL_SOCKET, SO_PASSCRED, true);
1712         if (r < 0)
1713                 return log_error_errno(r, "Failed to enable SO_PASSCRED: %m");
1714
1715         r = udev_watch_init();
1716         if (r < 0)
1717                 return log_error_errno(r, "Failed to create inotify descriptor: %m");
1718         manager->fd_inotify = r;
1719
1720         udev_watch_restore();
1721
1722         /* block and listen to all signals on signalfd */
1723         assert_se(sigprocmask_many(SIG_BLOCK, NULL, SIGTERM, SIGINT, SIGHUP, SIGCHLD, -1) >= 0);
1724
1725         r = sd_event_default(&manager->event);
1726         if (r < 0)
1727                 return log_error_errno(r, "Failed to allocate event loop: %m");
1728
1729         r = sd_event_add_signal(manager->event, NULL, SIGINT, on_sigterm, manager);
1730         if (r < 0)
1731                 return log_error_errno(r, "Failed to create SIGINT event source: %m");
1732
1733         r = sd_event_add_signal(manager->event, NULL, SIGTERM, on_sigterm, manager);
1734         if (r < 0)
1735                 return log_error_errno(r, "Failed to create SIGTERM event source: %m");
1736
1737         r = sd_event_add_signal(manager->event, NULL, SIGHUP, on_sighup, manager);
1738         if (r < 0)
1739                 return log_error_errno(r, "Failed to create SIGHUP event source: %m");
1740
1741         r = sd_event_add_signal(manager->event, NULL, SIGCHLD, on_sigchld, manager);
1742         if (r < 0)
1743                 return log_error_errno(r, "Failed to create SIGCHLD event source: %m");
1744
1745         r = sd_event_set_watchdog(manager->event, true);
1746         if (r < 0)
1747                 return log_error_errno(r, "Failed to create watchdog event source: %m");
1748
1749         r = udev_ctrl_attach_event(manager->ctrl, manager->event);
1750         if (r < 0)
1751                 return log_error_errno(r, "Failed to attach event to udev control: %m");
1752
1753         r = udev_ctrl_start(manager->ctrl, on_ctrl_msg, manager);
1754         if (r < 0)
1755                 return log_error_errno(r, "Failed to start device monitor: %m");
1756
1757         /* This needs to be after the inotify and uevent handling, to make sure
1758          * that the ping is send back after fully processing the pending uevents
1759          * (including the synthetic ones we may create due to inotify events).
1760          */
1761         r = sd_event_source_set_priority(udev_ctrl_get_event_source(manager->ctrl), SD_EVENT_PRIORITY_IDLE);
1762         if (r < 0)
1763                 return log_error_errno(r, "Failed to set IDLE event priority for udev control event source: %m");
1764
1765         r = sd_event_add_io(manager->event, &manager->inotify_event, manager->fd_inotify, EPOLLIN, on_inotify, manager);
1766         if (r < 0)
1767                 return log_error_errno(r, "Failed to create inotify event source: %m");
1768
1769         r = sd_device_monitor_attach_event(manager->monitor, manager->event);
1770         if (r < 0)
1771                 return log_error_errno(r, "Failed to attach event to device monitor: %m");
1772
1773         r = sd_device_monitor_start(manager->monitor, on_uevent, manager);
1774         if (r < 0)
1775                 return log_error_errno(r, "Failed to start device monitor: %m");
1776
1777         (void) sd_event_source_set_description(sd_device_monitor_get_event_source(manager->monitor), "device-monitor");
1778
1779         r = sd_event_add_io(manager->event, NULL, fd_worker, EPOLLIN, on_worker, manager);
1780         if (r < 0)
1781                 return log_error_errno(r, "Failed to create worker event source: %m");
1782
1783         r = sd_event_add_post(manager->event, NULL, on_post, manager);
1784         if (r < 0)
1785                 return log_error_errno(r, "Failed to create post event source: %m");
1786
1787         udev_builtin_init();
1788
1789         r = udev_rules_load(&manager->rules, arg_resolve_name_timing);
1790         if (!manager->rules)
1791                 return log_error_errno(r, "Failed to read udev rules: %m");
1792
1793         r = udev_rules_apply_static_dev_perms(manager->rules);
1794         if (r < 0)
1795                 log_error_errno(r, "Failed to apply permissions on static device nodes: %m");
1796
1797         (void) sd_notifyf(false,
1798                           "READY=1\n"
1799                           "STATUS=Processing with %u children at max", arg_children_max);
1800
1801         r = sd_event_loop(manager->event);
1802         if (r < 0)
1803                 log_error_errno(r, "Event loop failed: %m");
1804
1805         sd_notify(false,
1806                   "STOPPING=1\n"
1807                   "STATUS=Shutting down...");
1808         return r;
1809 }
1810
1811 int run_udevd(int argc, char *argv[]) {
1812         _cleanup_free_ char *cgroup = NULL;
1813         _cleanup_(manager_freep) Manager *manager = NULL;
1814         int fd_ctrl = -1, fd_uevent = -1;
1815         int r;
1816
1817         log_set_target(LOG_TARGET_AUTO);
1818         log_open();
1819         udev_parse_config_full(&arg_children_max, &arg_exec_delay_usec, &arg_event_timeout_usec, &arg_resolve_name_timing, &arg_timeout_signal);
1820         log_parse_environment();
1821         log_open(); /* Done again to update after reading configuration. */
1822
1823         r = parse_argv(argc, argv);
1824         if (r <= 0)
1825                 return r;
1826
1827         r = proc_cmdline_parse(parse_proc_cmdline_item, NULL, PROC_CMDLINE_STRIP_RD_PREFIX);
1828         if (r < 0)
1829                 log_warning_errno(r, "Failed to parse kernel command line, ignoring: %m");
1830
1831         if (arg_debug) {
1832                 log_set_target(LOG_TARGET_CONSOLE);
1833                 log_set_max_level(LOG_DEBUG);
1834         }
1835
1836         log_set_max_level_realm(LOG_REALM_SYSTEMD, log_get_max_level());
1837
1838         r = must_be_root();
1839         if (r < 0)
1840                 return r;
1841
1842         if (arg_children_max == 0) {
1843                 unsigned long cpu_limit, mem_limit, cpu_count = 1;
1844
1845                 r = cpus_in_affinity_mask();
1846                 if (r < 0)
1847                         log_warning_errno(r, "Failed to determine number of local CPUs, ignoring: %m");
1848                 else
1849                         cpu_count = r;
1850
1851                 cpu_limit = cpu_count * 2 + 16;
1852                 mem_limit = MAX(physical_memory() / (128UL*1024*1024), 10U);
1853
1854                 arg_children_max = MIN(cpu_limit, mem_limit);
1855                 arg_children_max = MIN(WORKER_NUM_MAX, arg_children_max);
1856
1857                 log_debug("Set children_max to %u", arg_children_max);
1858         }
1859
1860         /* set umask before creating any file/directory */
1861         umask(022);
1862
1863         r = mac_selinux_init();
1864         if (r < 0)
1865                 return r;
1866
1867         r = mkdir_errno_wrapper("/run/udev", 0755);
1868         if (r < 0 && r != -EEXIST)
1869                 return log_error_errno(r, "Failed to create /run/udev: %m");
1870
1871         if (getppid() == 1 && sd_booted() > 0) {
1872                 /* Get our own cgroup, we regularly kill everything udev has left behind.
1873                  * We only do this on systemd systems, and only if we are directly spawned
1874                  * by PID1. Otherwise we are not guaranteed to have a dedicated cgroup. */
1875                 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &cgroup);
1876                 if (r < 0) {
1877                         if (IN_SET(r, -ENOENT, -ENOMEDIUM))
1878                                 log_debug_errno(r, "Dedicated cgroup not found: %m");
1879                         else
1880                                 log_warning_errno(r, "Failed to get cgroup: %m");
1881                 }
1882         }
1883
1884         r = listen_fds(&fd_ctrl, &fd_uevent);
1885         if (r < 0)
1886                 return log_error_errno(r, "Failed to listen on fds: %m");
1887
1888         r = manager_new(&manager, fd_ctrl, fd_uevent, cgroup);
1889         if (r < 0)
1890                 return log_error_errno(r, "Failed to create manager: %m");
1891
1892         if (arg_daemonize) {
1893                 pid_t pid;
1894
1895                 log_info("Starting version " GIT_VERSION);
1896
1897                 /* connect /dev/null to stdin, stdout, stderr */
1898                 if (log_get_max_level() < LOG_DEBUG) {
1899                         r = make_null_stdio();
1900                         if (r < 0)
1901                                 log_warning_errno(r, "Failed to redirect standard streams to /dev/null: %m");
1902                 }
1903
1904                 pid = fork();
1905                 if (pid < 0)
1906                         return log_error_errno(errno, "Failed to fork daemon: %m");
1907                 if (pid > 0)
1908                         /* parent */
1909                         return 0;
1910
1911                 /* child */
1912                 (void) setsid();
1913         }
1914
1915         return main_loop(manager);
1916 }