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         int r;
 616
 617         assert(manager);
 618         assert(event);
 619
 620         if (DEBUG_LOGGING) {
 621                 DeviceAction action;
 622
 623                 r = device_get_action(event->dev, &action);
 624                 log_device_debug(event->dev, "Device (SEQNUM=%"PRIu64", ACTION=%s) ready for processing",
 625                                  event->seqnum, r >= 0 ? device_action_to_string(action) : "<unknown>");
 626         }
 627
 628         HASHMAP_FOREACH(worker, manager->workers) {
 629                 if (worker->state != WORKER_IDLE)
 630                         continue;
 631
 632                 r = device_monitor_send_device(manager->monitor, worker->monitor, event->dev);
 633                 if (r < 0) {
 634                         log_device_error_errno(event->dev, r, "Worker ["PID_FMT"] did not accept message, killing the worker: %m",
 635                                                worker->pid);
 636                         (void) kill(worker->pid, SIGKILL);
 637                         worker->state = WORKER_KILLED;
 638                         continue;
 639                 }
 640                 worker_attach_event(worker, event);
 641                 return;
 642         }
 643
 644         if (hashmap_size(manager->workers) >= arg_children_max) {
 645
 646                 /* Avoid spamming the debug logs if the limit is already reached and
 647                  * many events still need to be processed */
 648                 if (log_children_max_reached && arg_children_max > 1) {
 649                         log_debug("Maximum number (%u) of children reached.", hashmap_size(manager->workers));
 650                         log_children_max_reached = false;
 651                 }
 652                 return;
 653         }
 654
 655         /* Re-enable the debug message for the next batch of events */
 656         log_children_max_reached = true;
 657
 658         /* fork with up-to-date SELinux label database, so the child inherits the up-to-date db
 659            and, until the next SELinux policy changes, we safe further reloads in future children */
 660         mac_selinux_maybe_reload();
 661
 662         /* start new worker and pass initial device */
 663         worker_spawn(manager, event);
 664 }
 665
 666 static int event_queue_insert(Manager *manager, sd_device *dev) {
 667         _cleanup_(sd_device_unrefp) sd_device *clone = NULL;
 668         struct event *event;
 669         DeviceAction action;
 670         uint64_t seqnum;
 671         int r;
 672
 673         assert(manager);
 674         assert(dev);
 675
 676         /* only one process can add events to the queue */
 677         assert(manager->pid == getpid_cached());
 678
 679         /* We only accepts devices received by device monitor. */
 680         r = device_get_seqnum(dev, &seqnum);
 681         if (r < 0)
 682                 return r;
 683
 684         /* Refuse devices do not have ACTION property. */
 685         r = device_get_action(dev, &action);
 686         if (r < 0)
 687                 return r;
 688
 689         /* Save original device to restore the state on failures. */
 690         r = device_shallow_clone(dev, &clone);
 691         if (r < 0)
 692                 return r;
 693
 694         r = device_copy_properties(clone, dev);
 695         if (r < 0)
 696                 return r;
 697
 698         event = new(struct event, 1);
 699         if (!event)
 700                 return -ENOMEM;
 701
 702         *event = (struct event) {
 703                 .manager = manager,
 704                 .dev = sd_device_ref(dev),
 705                 .dev_kernel = TAKE_PTR(clone),
 706                 .seqnum = seqnum,
 707                 .state = EVENT_QUEUED,
 708         };
 709
 710         if (LIST_IS_EMPTY(manager->events)) {
 711                 r = touch("/run/udev/queue");
 712                 if (r < 0)
 713                         log_warning_errno(r, "Failed to touch /run/udev/queue: %m");
 714         }
 715
 716         LIST_APPEND(event, manager->events, event);
 717
 718         log_device_debug(dev, "Device (SEQNUM=%"PRIu64", ACTION=%s) is queued",
 719                          seqnum, device_action_to_string(action));
 720
 721         return 0;
 722 }
 723
 724 static void manager_kill_workers(Manager *manager) {
 725         struct worker *worker;
 726
 727         assert(manager);
 728
 729         HASHMAP_FOREACH(worker, manager->workers) {
 730                 if (worker->state == WORKER_KILLED)
 731                         continue;
 732
 733                 worker->state = WORKER_KILLED;
 734                 (void) kill(worker->pid, SIGTERM);
 735         }
 736 }
 737
 738 /* lookup event for identical, parent, child device */
 739 static int is_device_busy(Manager *manager, struct event *event) {
 740         const char *subsystem, *devpath, *devpath_old = NULL;
 741         dev_t devnum = makedev(0, 0);
 742         struct event *loop_event;
 743         size_t devpath_len;
 744         int r, ifindex = 0;
 745         bool is_block;
 746
 747         r = sd_device_get_subsystem(event->dev, &subsystem);
 748         if (r < 0)
 749                 return r;
 750
 751         is_block = streq(subsystem, "block");
 752
 753         r = sd_device_get_devpath(event->dev, &devpath);
 754         if (r < 0)
 755                 return r;
 756
 757         devpath_len = strlen(devpath);
 758
 759         r = sd_device_get_property_value(event->dev, "DEVPATH_OLD", &devpath_old);
 760         if (r < 0 && r != -ENOENT)
 761                 return r;
 762
 763         r = sd_device_get_devnum(event->dev, &devnum);
 764         if (r < 0 && r != -ENOENT)
 765                 return r;
 766
 767         r = sd_device_get_ifindex(event->dev, &ifindex);
 768         if (r < 0 && r != -ENOENT)
 769                 return r;
 770
 771         /* check if queue contains events we depend on */
 772         LIST_FOREACH(event, loop_event, manager->events) {
 773                 size_t loop_devpath_len, common;
 774                 const char *loop_devpath;
 775
 776                 /* we already found a later event, earlier cannot block us, no need to check again */
 777                 if (loop_event->seqnum < event->delaying_seqnum)
 778                         continue;
 779
 780                 /* event we checked earlier still exists, no need to check again */
 781                 if (loop_event->seqnum == event->delaying_seqnum)
 782                         return true;
 783
 784                 /* found ourself, no later event can block us */
 785                 if (loop_event->seqnum >= event->seqnum)
 786                         break;
 787
 788                 /* check major/minor */
 789                 if (major(devnum) != 0) {
 790                         const char *s;
 791                         dev_t d;
 792
 793                         if (sd_device_get_subsystem(loop_event->dev, &s) < 0)
 794                                 continue;
 795
 796                         if (sd_device_get_devnum(loop_event->dev, &d) >= 0 &&
 797                             devnum == d && is_block == streq(s, "block"))
 798                                 goto set_delaying_seqnum;
 799                 }
 800
 801                 /* check network device ifindex */
 802                 if (ifindex > 0) {
 803                         int i;
 804
 805                         if (sd_device_get_ifindex(loop_event->dev, &i) >= 0 &&
 806                             ifindex == i)
 807                                 goto set_delaying_seqnum;
 808                 }
 809
 810                 if (sd_device_get_devpath(loop_event->dev, &loop_devpath) < 0)
 811                         continue;
 812
 813                 /* check our old name */
 814                 if (devpath_old && streq(devpath_old, loop_devpath))
 815                         goto set_delaying_seqnum;
 816
 817                 loop_devpath_len = strlen(loop_devpath);
 818
 819                 /* compare devpath */
 820                 common = MIN(devpath_len, loop_devpath_len);
 821
 822                 /* one devpath is contained in the other? */
 823                 if (!strneq(devpath, loop_devpath, common))
 824                         continue;
 825
 826                 /* identical device event found */
 827                 if (devpath_len == loop_devpath_len)
 828                         goto set_delaying_seqnum;
 829
 830                 /* parent device event found */
 831                 if (devpath[common] == '/')
 832                         goto set_delaying_seqnum;
 833
 834                 /* child device event found */
 835                 if (loop_devpath[common] == '/')
 836                         goto set_delaying_seqnum;
 837         }
 838
 839         return false;
 840
 841 set_delaying_seqnum:
 842         log_device_debug(event->dev, "SEQNUM=%" PRIu64 " blocked by SEQNUM=%" PRIu64,
 843                          event->seqnum, loop_event->seqnum);
 844
 845         event->delaying_seqnum = loop_event->seqnum;
 846         return true;
 847 }
 848
 849 static void manager_exit(Manager *manager) {
 850         assert(manager);
 851
 852         manager->exit = true;
 853
 854         sd_notify(false,
 855                   "STOPPING=1\n"
 856                   "STATUS=Starting shutdown...");
 857
 858         /* close sources of new events and discard buffered events */
 859         manager->ctrl = udev_ctrl_unref(manager->ctrl);
 860
 861         manager->inotify_event = sd_event_source_unref(manager->inotify_event);
 862         manager->fd_inotify = safe_close(manager->fd_inotify);
 863
 864         manager->monitor = sd_device_monitor_unref(manager->monitor);
 865
 866         /* discard queued events and kill workers */
 867         event_queue_cleanup(manager, EVENT_QUEUED);
 868         manager_kill_workers(manager);
 869 }
 870
 871 /* reload requested, HUP signal received, rules changed, builtin changed */
 872 static void manager_reload(Manager *manager) {
 873
 874         assert(manager);
 875
 876         sd_notify(false,
 877                   "RELOADING=1\n"
 878                   "STATUS=Flushing configuration...");
 879
 880         manager_kill_workers(manager);
 881         manager->rules = udev_rules_free(manager->rules);
 882         udev_builtin_exit();
 883
 884         sd_notifyf(false,
 885                    "READY=1\n"
 886                    "STATUS=Processing with %u children at max", arg_children_max);
 887 }
 888
 889 static int on_kill_workers_event(sd_event_source *s, uint64_t usec, void *userdata) {
 890         Manager *manager = userdata;
 891
 892         assert(manager);
 893
 894         log_debug("Cleanup idle workers");
 895         manager_kill_workers(manager);
 896
 897         return 1;
 898 }
 899
 900 static void event_queue_start(Manager *manager) {
 901         struct event *event;
 902         usec_t usec;
 903         int r;
 904
 905         assert(manager);
 906
 907         if (LIST_IS_EMPTY(manager->events) ||
 908             manager->exit || manager->stop_exec_queue)
 909                 return;
 910
 911         assert_se(sd_event_now(manager->event, CLOCK_MONOTONIC, &usec) >= 0);
 912         /* check for changed config, every 3 seconds at most */
 913         if (manager->last_usec == 0 ||
 914             usec - manager->last_usec > 3 * USEC_PER_SEC) {
 915                 if (udev_rules_check_timestamp(manager->rules) ||
 916                     udev_builtin_validate())
 917                         manager_reload(manager);
 918
 919                 manager->last_usec = usec;
 920         }
 921
 922         r = event_source_disable(manager->kill_workers_event);
 923         if (r < 0)
 924                 log_warning_errno(r, "Failed to disable event source for cleaning up idle workers, ignoring: %m");
 925
 926         udev_builtin_init();
 927
 928         if (!manager->rules) {
 929                 r = udev_rules_load(&manager->rules, arg_resolve_name_timing);
 930                 if (r < 0) {
 931                         log_warning_errno(r, "Failed to read udev rules: %m");
 932                         return;
 933                 }
 934         }
 935
 936         LIST_FOREACH(event, event, manager->events) {
 937                 if (event->state != EVENT_QUEUED)
 938                         continue;
 939
 940                 /* do not start event if parent or child event is still running */
 941                 if (is_device_busy(manager, event) != 0)
 942                         continue;
 943
 944                 event_run(manager, event);
 945         }
 946 }
 947
 948 static void event_queue_cleanup(Manager *manager, enum event_state match_type) {
 949         struct event *event, *tmp;
 950
 951         LIST_FOREACH_SAFE(event, event, tmp, manager->events) {
 952                 if (match_type != EVENT_UNDEF && match_type != event->state)
 953                         continue;
 954
 955                 event_free(event);
 956         }
 957 }
 958
 959 static int on_worker(sd_event_source *s, int fd, uint32_t revents, void *userdata) {
 960         Manager *manager = userdata;
 961
 962         assert(manager);
 963
 964         for (;;) {
 965                 struct worker_message msg;
 966                 struct iovec iovec = {
 967                         .iov_base = &msg,
 968                         .iov_len = sizeof(msg),
 969                 };
 970                 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred))) control;
 971                 struct msghdr msghdr = {
 972                         .msg_iov = &iovec,
 973                         .msg_iovlen = 1,
 974                         .msg_control = &control,
 975                         .msg_controllen = sizeof(control),
 976                 };
 977                 ssize_t size;
 978                 struct ucred *ucred;
 979                 struct worker *worker;
 980
 981                 size = recvmsg_safe(fd, &msghdr, MSG_DONTWAIT);
 982                 if (size == -EINTR)
 983                         continue;
 984                 if (size == -EAGAIN)
 985                         /* nothing more to read */
 986                         break;
 987                 if (size < 0)
 988                         return log_error_errno(size, "Failed to receive message: %m");
 989
 990                 cmsg_close_all(&msghdr);
 991
 992                 if (size != sizeof(struct worker_message)) {
 993                         log_warning("Ignoring worker message with invalid size %zi bytes", size);
 994                         continue;
 995                 }
 996
 997                 ucred = CMSG_FIND_DATA(&msghdr, SOL_SOCKET, SCM_CREDENTIALS, struct ucred);
 998                 if (!ucred || ucred->pid <= 0) {
 999                         log_warning("Ignoring worker message without valid PID");
1000                         continue;
1001                 }
1002
1003                 /* lookup worker who sent the signal */
1004                 worker = hashmap_get(manager->workers, PID_TO_PTR(ucred->pid));
1005                 if (!worker) {
1006                         log_debug("Worker ["PID_FMT"] returned, but is no longer tracked", ucred->pid);
1007                         continue;
1008                 }
1009
1010                 if (worker->state != WORKER_KILLED)
1011                         worker->state = WORKER_IDLE;
1012
1013                 /* worker returned */
1014                 event_free(worker->event);
1015         }
1016
1017         /* we have free workers, try to schedule events */
1018         event_queue_start(manager);
1019
1020         return 1;
1021 }
1022
1023 static int on_uevent(sd_device_monitor *monitor, sd_device *dev, void *userdata) {
1024         Manager *manager = userdata;
1025         int r;
1026
1027         assert(manager);
1028
1029         device_ensure_usec_initialized(dev, NULL);
1030
1031         r = event_queue_insert(manager, dev);
1032         if (r < 0) {
1033                 log_device_error_errno(dev, r, "Failed to insert device into event queue: %m");
1034                 return 1;
1035         }
1036
1037         /* we have fresh events, try to schedule them */
1038         event_queue_start(manager);
1039
1040         return 1;
1041 }
1042
1043 /* receive the udevd message from userspace */
1044 static int on_ctrl_msg(struct udev_ctrl *uctrl, enum udev_ctrl_msg_type type, const union udev_ctrl_msg_value *value, void *userdata) {
1045         Manager *manager = userdata;
1046         int r;
1047
1048         assert(value);
1049         assert(manager);
1050
1051         switch (type) {
1052         case UDEV_CTRL_SET_LOG_LEVEL:
1053                 log_debug("Received udev control message (SET_LOG_LEVEL), setting log_priority=%i", value->intval);
1054                 log_set_max_level_realm(LOG_REALM_UDEV, value->intval);
1055                 log_set_max_level_realm(LOG_REALM_SYSTEMD, value->intval);
1056                 manager_kill_workers(manager);
1057                 break;
1058         case UDEV_CTRL_STOP_EXEC_QUEUE:
1059                 log_debug("Received udev control message (STOP_EXEC_QUEUE)");
1060                 manager->stop_exec_queue = true;
1061                 break;
1062         case UDEV_CTRL_START_EXEC_QUEUE:
1063                 log_debug("Received udev control message (START_EXEC_QUEUE)");
1064                 manager->stop_exec_queue = false;
1065                 event_queue_start(manager);
1066                 break;
1067         case UDEV_CTRL_RELOAD:
1068                 log_debug("Received udev control message (RELOAD)");
1069                 manager_reload(manager);
1070                 break;
1071         case UDEV_CTRL_SET_ENV: {
1072                 _cleanup_free_ char *key = NULL, *val = NULL, *old_key = NULL, *old_val = NULL;
1073                 const char *eq;
1074
1075                 eq = strchr(value->buf, '=');
1076                 if (!eq) {
1077                         log_error("Invalid key format '%s'", value->buf);
1078                         return 1;
1079                 }
1080
1081                 key = strndup(value->buf, eq - value->buf);
1082                 if (!key) {
1083                         log_oom();
1084                         return 1;
1085                 }
1086
1087                 old_val = hashmap_remove2(manager->properties, key, (void **) &old_key);
1088
1089                 r = hashmap_ensure_allocated(&manager->properties, &string_hash_ops);
1090                 if (r < 0) {
1091                         log_oom();
1092                         return 1;
1093                 }
1094
1095                 eq++;
1096                 if (isempty(eq)) {
1097                         log_debug("Received udev control message (ENV), unsetting '%s'", key);
1098
1099                         r = hashmap_put(manager->properties, key, NULL);
1100                         if (r < 0) {
1101                                 log_oom();
1102                                 return 1;
1103                         }
1104                 } else {
1105                         val = strdup(eq);
1106                         if (!val) {
1107                                 log_oom();
1108                                 return 1;
1109                         }
1110
1111                         log_debug("Received udev control message (ENV), setting '%s=%s'", key, val);
1112
1113                         r = hashmap_put(manager->properties, key, val);
1114                         if (r < 0) {
1115                                 log_oom();
1116                                 return 1;
1117                         }
1118                 }
1119
1120                 key = val = NULL;
1121                 manager_kill_workers(manager);
1122                 break;
1123         }
1124         case UDEV_CTRL_SET_CHILDREN_MAX:
1125                 if (value->intval <= 0) {
1126                         log_debug("Received invalid udev control message (SET_MAX_CHILDREN, %i), ignoring.", value->intval);
1127                         return 0;
1128                 }
1129
1130                 log_debug("Received udev control message (SET_MAX_CHILDREN), setting children_max=%i", value->intval);
1131                 arg_children_max = value->intval;
1132
1133                 (void) sd_notifyf(false,
1134                                   "READY=1\n"
1135                                   "STATUS=Processing with %u children at max", arg_children_max);
1136                 break;
1137         case UDEV_CTRL_PING:
1138                 log_debug("Received udev control message (PING)");
1139                 break;
1140         case UDEV_CTRL_EXIT:
1141                 log_debug("Received udev control message (EXIT)");
1142                 manager_exit(manager);
1143                 break;
1144         default:
1145                 log_debug("Received unknown udev control message, ignoring");
1146         }
1147
1148         return 1;
1149 }
1150
1151 static int synthesize_change_one(sd_device *dev, const char *syspath) {
1152         const char *filename;
1153         int r;
1154
1155         filename = strjoina(syspath, "/uevent");
1156         log_device_debug(dev, "device is closed, synthesising 'change' on %s", syspath);
1157         r = write_string_file(filename, "change", WRITE_STRING_FILE_DISABLE_BUFFER);
1158         if (r < 0)
1159                 return log_device_debug_errno(dev, r, "Failed to write 'change' to %s: %m", filename);
1160         return 0;
1161 }
1162
1163 static int synthesize_change(sd_device *dev) {
1164         const char *subsystem, *sysname, *devname, *syspath, *devtype;
1165         int r;
1166
1167         r = sd_device_get_subsystem(dev, &subsystem);
1168         if (r < 0)
1169                 return r;
1170
1171         r = sd_device_get_sysname(dev, &sysname);
1172         if (r < 0)
1173                 return r;
1174
1175         r = sd_device_get_devname(dev, &devname);
1176         if (r < 0)
1177                 return r;
1178
1179         r = sd_device_get_syspath(dev, &syspath);
1180         if (r < 0)
1181                 return r;
1182
1183         r = sd_device_get_devtype(dev, &devtype);
1184         if (r < 0)
1185                 return r;
1186
1187         if (streq_ptr("block", subsystem) &&
1188             streq_ptr("disk", devtype) &&
1189             !startswith(sysname, "dm-")) {
1190                 _cleanup_(sd_device_enumerator_unrefp) sd_device_enumerator *e = NULL;
1191                 bool part_table_read = false, has_partitions = false;
1192                 sd_device *d;
1193                 int fd;
1194
1195                 /*
1196                  * Try to re-read the partition table. This only succeeds if
1197                  * none of the devices is busy. The kernel returns 0 if no
1198                  * partition table is found, and we will not get an event for
1199                  * the disk.
1200                  */
1201                 fd = open(devname, O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
1202                 if (fd >= 0) {
1203                         r = flock(fd, LOCK_EX|LOCK_NB);
1204                         if (r >= 0)
1205                                 r = ioctl(fd, BLKRRPART, 0);
1206
1207                         close(fd);
1208                         if (r >= 0)
1209                                 part_table_read = true;
1210                 }
1211
1212                 /* search for partitions */
1213                 r = sd_device_enumerator_new(&e);
1214                 if (r < 0)
1215                         return r;
1216
1217                 r = sd_device_enumerator_allow_uninitialized(e);
1218                 if (r < 0)
1219                         return r;
1220
1221                 r = sd_device_enumerator_add_match_parent(e, dev);
1222                 if (r < 0)
1223                         return r;
1224
1225                 r = sd_device_enumerator_add_match_subsystem(e, "block", true);
1226                 if (r < 0)
1227                         return r;
1228
1229                 FOREACH_DEVICE(e, d) {
1230                         const char *t;
1231
1232                         if (sd_device_get_devtype(d, &t) < 0 ||
1233                             !streq("partition", t))
1234                                 continue;
1235
1236                         has_partitions = true;
1237                         break;
1238                 }
1239
1240                 /*
1241                  * We have partitions and re-read the table, the kernel already sent
1242                  * out a "change" event for the disk, and "remove/add" for all
1243                  * partitions.
1244                  */
1245                 if (part_table_read && has_partitions)
1246                         return 0;
1247
1248                 /*
1249                  * We have partitions but re-reading the partition table did not
1250                  * work, synthesize "change" for the disk and all partitions.
1251                  */
1252                 (void) synthesize_change_one(dev, syspath);
1253
1254                 FOREACH_DEVICE(e, d) {
1255                         const char *t, *n, *s;
1256
1257                         if (sd_device_get_devtype(d, &t) < 0 ||
1258                             !streq("partition", t))
1259                                 continue;
1260
1261                         if (sd_device_get_devname(d, &n) < 0 ||
1262                             sd_device_get_syspath(d, &s) < 0)
1263                                 continue;
1264
1265                         (void) synthesize_change_one(dev, s);
1266                 }
1267
1268         } else
1269                 (void) synthesize_change_one(dev, syspath);
1270
1271         return 0;
1272 }
1273
1274 static int on_inotify(sd_event_source *s, int fd, uint32_t revents, void *userdata) {
1275         Manager *manager = userdata;
1276         union inotify_event_buffer buffer;
1277         struct inotify_event *e;
1278         ssize_t l;
1279         int r;
1280
1281         assert(manager);
1282
1283         r = event_source_disable(manager->kill_workers_event);
1284         if (r < 0)
1285                 log_warning_errno(r, "Failed to disable event source for cleaning up idle workers, ignoring: %m");
1286
1287         l = read(fd, &buffer, sizeof(buffer));
1288         if (l < 0) {
1289                 if (IN_SET(errno, EAGAIN, EINTR))
1290                         return 1;
1291
1292                 return log_error_errno(errno, "Failed to read inotify fd: %m");
1293         }
1294
1295         FOREACH_INOTIFY_EVENT(e, buffer, l) {
1296                 _cleanup_(sd_device_unrefp) sd_device *dev = NULL;
1297                 const char *devnode;
1298
1299                 if (udev_watch_lookup(e->wd, &dev) <= 0)
1300                         continue;
1301
1302                 if (sd_device_get_devname(dev, &devnode) < 0)
1303                         continue;
1304
1305                 log_device_debug(dev, "Inotify event: %x for %s", e->mask, devnode);
1306                 if (e->mask & IN_CLOSE_WRITE)
1307                         synthesize_change(dev);
1308                 else if (e->mask & IN_IGNORED)
1309                         udev_watch_end(dev);
1310         }
1311
1312         return 1;
1313 }
1314
1315 static int on_sigterm(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
1316         Manager *manager = userdata;
1317
1318         assert(manager);
1319
1320         manager_exit(manager);
1321
1322         return 1;
1323 }
1324
1325 static int on_sighup(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
1326         Manager *manager = userdata;
1327
1328         assert(manager);
1329
1330         manager_reload(manager);
1331
1332         return 1;
1333 }
1334
1335 static int on_sigchld(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
1336         Manager *manager = userdata;
1337         int r;
1338
1339         assert(manager);
1340
1341         for (;;) {
1342                 pid_t pid;
1343                 int status;
1344                 struct worker *worker;
1345
1346                 pid = waitpid(-1, &status, WNOHANG);
1347                 if (pid <= 0)
1348                         break;
1349
1350                 worker = hashmap_get(manager->workers, PID_TO_PTR(pid));
1351                 if (!worker) {
1352                         log_warning("Worker ["PID_FMT"] is unknown, ignoring", pid);
1353                         continue;
1354                 }
1355
1356                 if (WIFEXITED(status)) {
1357                         if (WEXITSTATUS(status) == 0)
1358                                 log_debug("Worker ["PID_FMT"] exited", pid);
1359                         else
1360                                 log_warning("Worker ["PID_FMT"] exited with return code %i", pid, WEXITSTATUS(status));
1361                 } else if (WIFSIGNALED(status))
1362                         log_warning("Worker ["PID_FMT"] terminated by signal %i (%s)", pid, WTERMSIG(status), signal_to_string(WTERMSIG(status)));
1363                 else if (WIFSTOPPED(status)) {
1364                         log_info("Worker ["PID_FMT"] stopped", pid);
1365                         continue;
1366                 } else if (WIFCONTINUED(status)) {
1367                         log_info("Worker ["PID_FMT"] continued", pid);
1368                         continue;
1369                 } else
1370                         log_warning("Worker ["PID_FMT"] exit with status 0x%04x", pid, status);
1371
1372                 if ((!WIFEXITED(status) || WEXITSTATUS(status) != 0) && worker->event) {
1373                         log_device_error(worker->event->dev, "Worker ["PID_FMT"] failed", pid);
1374
1375                         /* delete state from disk */
1376                         device_delete_db(worker->event->dev);
1377                         device_tag_index(worker->event->dev, NULL, false);
1378
1379                         if (manager->monitor) {
1380                                 /* forward kernel event without amending it */
1381                                 r = device_monitor_send_device(manager->monitor, NULL, worker->event->dev_kernel);
1382                                 if (r < 0)
1383                                         log_device_error_errno(worker->event->dev_kernel, r, "Failed to send back device to kernel: %m");
1384                         }
1385                 }
1386
1387                 worker_free(worker);
1388         }
1389
1390         /* we can start new workers, try to schedule events */
1391         event_queue_start(manager);
1392
1393         /* Disable unnecessary cleanup event */
1394         if (hashmap_isempty(manager->workers)) {
1395                 r = event_source_disable(manager->kill_workers_event);
1396                 if (r < 0)
1397                         log_warning_errno(r, "Failed to disable event source for cleaning up idle workers, ignoring: %m");
1398         }
1399
1400         return 1;
1401 }
1402
1403 static int on_post(sd_event_source *s, void *userdata) {
1404         Manager *manager = userdata;
1405
1406         assert(manager);
1407
1408         if (!LIST_IS_EMPTY(manager->events))
1409                 return 1;
1410
1411         /* There are no pending events. Let's cleanup idle process. */
1412
1413         if (!hashmap_isempty(manager->workers)) {
1414                 /* There are idle workers */
1415                 (void) event_reset_time(manager->event, &manager->kill_workers_event, CLOCK_MONOTONIC,
1416                                         now(CLOCK_MONOTONIC) + 3 * USEC_PER_SEC, USEC_PER_SEC,
1417                                         on_kill_workers_event, manager, 0, "kill-workers-event", false);
1418                 return 1;
1419         }
1420
1421         /* There are no idle workers. */
1422
1423         if (manager->exit)
1424                 return sd_event_exit(manager->event, 0);
1425
1426         if (manager->cgroup)
1427                 /* cleanup possible left-over processes in our cgroup */
1428                 (void) cg_kill(SYSTEMD_CGROUP_CONTROLLER, manager->cgroup, SIGKILL, CGROUP_IGNORE_SELF, NULL, NULL, NULL);
1429
1430         return 1;
1431 }
1432
1433 static int listen_fds(int *ret_ctrl, int *ret_netlink) {
1434         int ctrl_fd = -1, netlink_fd = -1;
1435         int fd, n;
1436
1437         assert(ret_ctrl);
1438         assert(ret_netlink);
1439
1440         n = sd_listen_fds(true);
1441         if (n < 0)
1442                 return n;
1443
1444         for (fd = SD_LISTEN_FDS_START; fd < n + SD_LISTEN_FDS_START; fd++) {
1445                 if (sd_is_socket(fd, AF_LOCAL, SOCK_SEQPACKET, -1) > 0) {
1446                         if (ctrl_fd >= 0)
1447                                 return -EINVAL;
1448                         ctrl_fd = fd;
1449                         continue;
1450                 }
1451
1452                 if (sd_is_socket(fd, AF_NETLINK, SOCK_RAW, -1) > 0) {
1453                         if (netlink_fd >= 0)
1454                                 return -EINVAL;
1455                         netlink_fd = fd;
1456                         continue;
1457                 }
1458
1459                 return -EINVAL;
1460         }
1461
1462         *ret_ctrl = ctrl_fd;
1463         *ret_netlink = netlink_fd;
1464
1465         return 0;
1466 }
1467
1468 /*
1469  * read the kernel command line, in case we need to get into debug mode
1470  *   udev.log_priority=<level>                 syslog priority
1471  *   udev.children_max=<number of workers>     events are fully serialized if set to 1
1472  *   udev.exec_delay=<number of seconds>       delay execution of every executed program
1473  *   udev.event_timeout=<number of seconds>    seconds to wait before terminating an event
1474  *   udev.blockdev_read_only<=bool>            mark all block devices read-only when they appear
1475  */
1476 static int parse_proc_cmdline_item(const char *key, const char *value, void *data) {
1477         int r;
1478
1479         assert(key);
1480
1481         if (proc_cmdline_key_streq(key, "udev.log_priority")) {
1482
1483                 if (proc_cmdline_value_missing(key, value))
1484                         return 0;
1485
1486                 r = log_level_from_string(value);
1487                 if (r >= 0)
1488                         log_set_max_level(r);
1489
1490         } else if (proc_cmdline_key_streq(key, "udev.event_timeout")) {
1491
1492                 if (proc_cmdline_value_missing(key, value))
1493                         return 0;
1494
1495                 r = parse_sec(value, &arg_event_timeout_usec);
1496
1497         } else if (proc_cmdline_key_streq(key, "udev.children_max")) {
1498
1499                 if (proc_cmdline_value_missing(key, value))
1500                         return 0;
1501
1502                 r = safe_atou(value, &arg_children_max);
1503
1504         } else if (proc_cmdline_key_streq(key, "udev.exec_delay")) {
1505
1506                 if (proc_cmdline_value_missing(key, value))
1507                         return 0;
1508
1509                 r = parse_sec(value, &arg_exec_delay_usec);
1510
1511         } else if (proc_cmdline_key_streq(key, "udev.timeout_signal")) {
1512
1513                 if (proc_cmdline_value_missing(key, value))
1514                         return 0;
1515
1516                 r = signal_from_string(value);
1517                 if (r > 0)
1518                         arg_timeout_signal = r;
1519
1520         } else if (proc_cmdline_key_streq(key, "udev.blockdev_read_only")) {
1521
1522                 if (!value)
1523                         arg_blockdev_read_only = true;
1524                 else {
1525                         r = parse_boolean(value);
1526                         if (r < 0)
1527                                 log_warning_errno(r, "Failed to parse udev.blockdev-read-only argument, ignoring: %s", value);
1528                         else
1529                                 arg_blockdev_read_only = r;
1530                 }
1531
1532                 if (arg_blockdev_read_only)
1533                         log_notice("All physical block devices will be marked read-only.");
1534
1535                 return 0;
1536
1537         } else {
1538                 if (startswith(key, "udev."))
1539                         log_warning("Unknown udev kernel command line option \"%s\", ignoring.", key);
1540
1541                 return 0;
1542         }
1543
1544         if (r < 0)
1545                 log_warning_errno(r, "Failed to parse \"%s=%s\", ignoring: %m", key, value);
1546
1547         return 0;
1548 }
1549
1550 static int help(void) {
1551         _cleanup_free_ char *link = NULL;
1552         int r;
1553
1554         r = terminal_urlify_man("systemd-udevd.service", "8", &link);
1555         if (r < 0)
1556                 return log_oom();
1557
1558         printf("%s [OPTIONS...]\n\n"
1559                "Rule-based manager for device events and files.\n\n"
1560                "  -h --help                   Print this message\n"
1561                "  -V --version                Print version of the program\n"
1562                "  -d --daemon                 Detach and run in the background\n"
1563                "  -D --debug                  Enable debug output\n"
1564                "  -c --children-max=INT       Set maximum number of workers\n"
1565                "  -e --exec-delay=SECONDS     Seconds to wait before executing RUN=\n"
1566                "  -t --event-timeout=SECONDS  Seconds to wait before terminating an event\n"
1567                "  -N --resolve-names=early|late|never\n"
1568                "                              When to resolve users and groups\n"
1569                "\nSee the %s for details.\n"
1570                , program_invocation_short_name
1571                , link
1572         );
1573
1574         return 0;
1575 }
1576
1577 static int parse_argv(int argc, char *argv[]) {
1578         enum {
1579                 ARG_TIMEOUT_SIGNAL,
1580         };
1581
1582         static const struct option options[] = {
1583                 { "daemon",             no_argument,            NULL, 'd'                 },
1584                 { "debug",              no_argument,            NULL, 'D'                 },
1585                 { "children-max",       required_argument,      NULL, 'c'                 },
1586                 { "exec-delay",         required_argument,      NULL, 'e'                 },
1587                 { "event-timeout",      required_argument,      NULL, 't'                 },
1588                 { "resolve-names",      required_argument,      NULL, 'N'                 },
1589                 { "help",               no_argument,            NULL, 'h'                 },
1590                 { "version",            no_argument,            NULL, 'V'                 },
1591                 { "timeout-signal",     required_argument,      NULL,  ARG_TIMEOUT_SIGNAL },
1592                 {}
1593         };
1594
1595         int c, r;
1596
1597         assert(argc >= 0);
1598         assert(argv);
1599
1600         while ((c = getopt_long(argc, argv, "c:de:Dt:N:hV", options, NULL)) >= 0) {
1601                 switch (c) {
1602
1603                 case 'd':
1604                         arg_daemonize = true;
1605                         break;
1606                 case 'c':
1607                         r = safe_atou(optarg, &arg_children_max);
1608                         if (r < 0)
1609                                 log_warning_errno(r, "Failed to parse --children-max= value '%s', ignoring: %m", optarg);
1610                         break;
1611                 case 'e':
1612                         r = parse_sec(optarg, &arg_exec_delay_usec);
1613                         if (r < 0)
1614                                 log_warning_errno(r, "Failed to parse --exec-delay= value '%s', ignoring: %m", optarg);
1615                         break;
1616                 case ARG_TIMEOUT_SIGNAL:
1617                         r = signal_from_string(optarg);
1618                         if (r <= 0)
1619                                 log_warning_errno(r, "Failed to parse --timeout-signal= value '%s', ignoring: %m", optarg);
1620                         else
1621                                 arg_timeout_signal = r;
1622
1623                         break;
1624                 case 't':
1625                         r = parse_sec(optarg, &arg_event_timeout_usec);
1626                         if (r < 0)
1627                                 log_warning_errno(r, "Failed to parse --event-timeout= value '%s', ignoring: %m", optarg);
1628                         break;
1629                 case 'D':
1630                         arg_debug = true;
1631                         break;
1632                 case 'N': {
1633                         ResolveNameTiming t;
1634
1635                         t = resolve_name_timing_from_string(optarg);
1636                         if (t < 0)
1637                                 log_warning("Invalid --resolve-names= value '%s', ignoring.", optarg);
1638                         else
1639                                 arg_resolve_name_timing = t;
1640                         break;
1641                 }
1642                 case 'h':
1643                         return help();
1644                 case 'V':
1645                         printf("%s\n", GIT_VERSION);
1646                         return 0;
1647                 case '?':
1648                         return -EINVAL;
1649                 default:
1650                         assert_not_reached("Unhandled option");
1651
1652                 }
1653         }
1654
1655         return 1;
1656 }
1657
1658 static int manager_new(Manager **ret, int fd_ctrl, int fd_uevent, const char *cgroup) {
1659         _cleanup_(manager_freep) Manager *manager = NULL;
1660         int r;
1661
1662         assert(ret);
1663
1664         manager = new(Manager, 1);
1665         if (!manager)
1666                 return log_oom();
1667
1668         *manager = (Manager) {
1669                 .fd_inotify = -1,
1670                 .worker_watch = { -1, -1 },
1671                 .cgroup = cgroup,
1672         };
1673
1674         r = udev_ctrl_new_from_fd(&manager->ctrl, fd_ctrl);
1675         if (r < 0)
1676                 return log_error_errno(r, "Failed to initialize udev control socket: %m");
1677
1678         r = udev_ctrl_enable_receiving(manager->ctrl);
1679         if (r < 0)
1680                 return log_error_errno(r, "Failed to bind udev control socket: %m");
1681
1682         r = device_monitor_new_full(&manager->monitor, MONITOR_GROUP_KERNEL, fd_uevent);
1683         if (r < 0)
1684                 return log_error_errno(r, "Failed to initialize device monitor: %m");
1685
1686         /* Bump receiver buffer, but only if we are not called via socket activation, as in that
1687          * case systemd sets the receive buffer size for us, and the value in the .socket unit
1688          * should take full effect. */
1689         if (fd_uevent < 0)
1690                 (void) sd_device_monitor_set_receive_buffer_size(manager->monitor, 128 * 1024 * 1024);
1691
1692         r = device_monitor_enable_receiving(manager->monitor);
1693         if (r < 0)
1694                 return log_error_errno(r, "Failed to bind netlink socket: %m");
1695
1696         *ret = TAKE_PTR(manager);
1697
1698         return 0;
1699 }
1700
1701 static int main_loop(Manager *manager) {
1702         int fd_worker, r;
1703
1704         manager->pid = getpid_cached();
1705
1706         /* unnamed socket from workers to the main daemon */
1707         r = socketpair(AF_LOCAL, SOCK_DGRAM|SOCK_CLOEXEC, 0, manager->worker_watch);
1708         if (r < 0)
1709                 return log_error_errno(errno, "Failed to create socketpair for communicating with workers: %m");
1710
1711         fd_worker = manager->worker_watch[READ_END];
1712
1713         r = setsockopt_int(fd_worker, SOL_SOCKET, SO_PASSCRED, true);
1714         if (r < 0)
1715                 return log_error_errno(r, "Failed to enable SO_PASSCRED: %m");
1716
1717         r = udev_watch_init();
1718         if (r < 0)
1719                 return log_error_errno(r, "Failed to create inotify descriptor: %m");
1720         manager->fd_inotify = r;
1721
1722         udev_watch_restore();
1723
1724         /* block and listen to all signals on signalfd */
1725         assert_se(sigprocmask_many(SIG_BLOCK, NULL, SIGTERM, SIGINT, SIGHUP, SIGCHLD, -1) >= 0);
1726
1727         r = sd_event_default(&manager->event);
1728         if (r < 0)
1729                 return log_error_errno(r, "Failed to allocate event loop: %m");
1730
1731         r = sd_event_add_signal(manager->event, NULL, SIGINT, on_sigterm, manager);
1732         if (r < 0)
1733                 return log_error_errno(r, "Failed to create SIGINT event source: %m");
1734
1735         r = sd_event_add_signal(manager->event, NULL, SIGTERM, on_sigterm, manager);
1736         if (r < 0)
1737                 return log_error_errno(r, "Failed to create SIGTERM event source: %m");
1738
1739         r = sd_event_add_signal(manager->event, NULL, SIGHUP, on_sighup, manager);
1740         if (r < 0)
1741                 return log_error_errno(r, "Failed to create SIGHUP event source: %m");
1742
1743         r = sd_event_add_signal(manager->event, NULL, SIGCHLD, on_sigchld, manager);
1744         if (r < 0)
1745                 return log_error_errno(r, "Failed to create SIGCHLD event source: %m");
1746
1747         r = sd_event_set_watchdog(manager->event, true);
1748         if (r < 0)
1749                 return log_error_errno(r, "Failed to create watchdog event source: %m");
1750
1751         r = udev_ctrl_attach_event(manager->ctrl, manager->event);
1752         if (r < 0)
1753                 return log_error_errno(r, "Failed to attach event to udev control: %m");
1754
1755         r = udev_ctrl_start(manager->ctrl, on_ctrl_msg, manager);
1756         if (r < 0)
1757                 return log_error_errno(r, "Failed to start device monitor: %m");
1758
1759         /* This needs to be after the inotify and uevent handling, to make sure
1760          * that the ping is send back after fully processing the pending uevents
1761          * (including the synthetic ones we may create due to inotify events).
1762          */
1763         r = sd_event_source_set_priority(udev_ctrl_get_event_source(manager->ctrl), SD_EVENT_PRIORITY_IDLE);
1764         if (r < 0)
1765                 return log_error_errno(r, "Failed to set IDLE event priority for udev control event source: %m");
1766
1767         r = sd_event_add_io(manager->event, &manager->inotify_event, manager->fd_inotify, EPOLLIN, on_inotify, manager);
1768         if (r < 0)
1769                 return log_error_errno(r, "Failed to create inotify event source: %m");
1770
1771         r = sd_device_monitor_attach_event(manager->monitor, manager->event);
1772         if (r < 0)
1773                 return log_error_errno(r, "Failed to attach event to device monitor: %m");
1774
1775         r = sd_device_monitor_start(manager->monitor, on_uevent, manager);
1776         if (r < 0)
1777                 return log_error_errno(r, "Failed to start device monitor: %m");
1778
1779         (void) sd_event_source_set_description(sd_device_monitor_get_event_source(manager->monitor), "device-monitor");
1780
1781         r = sd_event_add_io(manager->event, NULL, fd_worker, EPOLLIN, on_worker, manager);
1782         if (r < 0)
1783                 return log_error_errno(r, "Failed to create worker event source: %m");
1784
1785         r = sd_event_add_post(manager->event, NULL, on_post, manager);
1786         if (r < 0)
1787                 return log_error_errno(r, "Failed to create post event source: %m");
1788
1789         udev_builtin_init();
1790
1791         r = udev_rules_load(&manager->rules, arg_resolve_name_timing);
1792         if (!manager->rules)
1793                 return log_error_errno(r, "Failed to read udev rules: %m");
1794
1795         r = udev_rules_apply_static_dev_perms(manager->rules);
1796         if (r < 0)
1797                 log_error_errno(r, "Failed to apply permissions on static device nodes: %m");
1798
1799         (void) sd_notifyf(false,
1800                           "READY=1\n"
1801                           "STATUS=Processing with %u children at max", arg_children_max);
1802
1803         r = sd_event_loop(manager->event);
1804         if (r < 0)
1805                 log_error_errno(r, "Event loop failed: %m");
1806
1807         sd_notify(false,
1808                   "STOPPING=1\n"
1809                   "STATUS=Shutting down...");
1810         return r;
1811 }
1812
1813 int run_udevd(int argc, char *argv[]) {
1814         _cleanup_free_ char *cgroup = NULL;
1815         _cleanup_(manager_freep) Manager *manager = NULL;
1816         int fd_ctrl = -1, fd_uevent = -1;
1817         int r;
1818
1819         log_set_target(LOG_TARGET_AUTO);
1820         log_open();
1821         udev_parse_config_full(&arg_children_max, &arg_exec_delay_usec, &arg_event_timeout_usec, &arg_resolve_name_timing, &arg_timeout_signal);
1822         log_parse_environment();
1823         log_open(); /* Done again to update after reading configuration. */
1824
1825         r = parse_argv(argc, argv);
1826         if (r <= 0)
1827                 return r;
1828
1829         r = proc_cmdline_parse(parse_proc_cmdline_item, NULL, PROC_CMDLINE_STRIP_RD_PREFIX);
1830         if (r < 0)
1831                 log_warning_errno(r, "Failed to parse kernel command line, ignoring: %m");
1832
1833         if (arg_debug) {
1834                 log_set_target(LOG_TARGET_CONSOLE);
1835                 log_set_max_level(LOG_DEBUG);
1836         }
1837
1838         log_set_max_level_realm(LOG_REALM_SYSTEMD, log_get_max_level());
1839
1840         r = must_be_root();
1841         if (r < 0)
1842                 return r;
1843
1844         if (arg_children_max == 0) {
1845                 unsigned long cpu_limit, mem_limit, cpu_count = 1;
1846
1847                 r = cpus_in_affinity_mask();
1848                 if (r < 0)
1849                         log_warning_errno(r, "Failed to determine number of local CPUs, ignoring: %m");
1850                 else
1851                         cpu_count = r;
1852
1853                 cpu_limit = cpu_count * 2 + 16;
1854                 mem_limit = MAX(physical_memory() / (128UL*1024*1024), 10U);
1855
1856                 arg_children_max = MIN(cpu_limit, mem_limit);
1857                 arg_children_max = MIN(WORKER_NUM_MAX, arg_children_max);
1858
1859                 log_debug("Set children_max to %u", arg_children_max);
1860         }
1861
1862         /* set umask before creating any file/directory */
1863         umask(022);
1864
1865         r = mac_selinux_init();
1866         if (r < 0)
1867                 return r;
1868
1869         r = mkdir_errno_wrapper("/run/udev", 0755);
1870         if (r < 0 && r != -EEXIST)
1871                 return log_error_errno(r, "Failed to create /run/udev: %m");
1872
1873         if (getppid() == 1 && sd_booted() > 0) {
1874                 /* Get our own cgroup, we regularly kill everything udev has left behind.
1875                  * We only do this on systemd systems, and only if we are directly spawned
1876                  * by PID1. Otherwise we are not guaranteed to have a dedicated cgroup. */
1877                 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &cgroup);
1878                 if (r < 0) {
1879                         if (IN_SET(r, -ENOENT, -ENOMEDIUM))
1880                                 log_debug_errno(r, "Dedicated cgroup not found: %m");
1881                         else
1882                                 log_warning_errno(r, "Failed to get cgroup: %m");
1883                 }
1884         }
1885
1886         r = listen_fds(&fd_ctrl, &fd_uevent);
1887         if (r < 0)
1888                 return log_error_errno(r, "Failed to listen on fds: %m");
1889
1890         r = manager_new(&manager, fd_ctrl, fd_uevent, cgroup);
1891         if (r < 0)
1892                 return log_error_errno(r, "Failed to create manager: %m");
1893
1894         if (arg_daemonize) {
1895                 pid_t pid;
1896
1897                 log_info("Starting version " GIT_VERSION);
1898
1899                 /* connect /dev/null to stdin, stdout, stderr */
1900                 if (log_get_max_level() < LOG_DEBUG) {
1901                         r = make_null_stdio();
1902                         if (r < 0)
1903                                 log_warning_errno(r, "Failed to redirect standard streams to /dev/null: %m");
1904                 }
1905
1906                 pid = fork();
1907                 if (pid < 0)
1908                         return log_error_errno(errno, "Failed to fork daemon: %m");
1909                 if (pid > 0)
1910                         /* parent */
1911                         return 0;
1912
1913                 /* child */
1914                 (void) setsid();
1915         }
1916
1917         return main_loop(manager);
1918 }