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 == -EAGAIN) {
 463                 /* So this is a block device and the device is locked currently via the BSD advisory locks —
 464                  * someone else is exclusively using it. This means we don't run our udev rules now, to not
 465                  * interfere. However we want to know when the device is unlocked again, and retrigger the
 466                  * device again then, so that the rules are run eventually. For that we use IN_CLOSE_WRITE
 467                  * inotify watches (which isn't exactly the same as waiting for the BSD locks to release, but
 468                  * not totally off, as long as unlock+close() is done together, as it usually is).
 469                  *
 470                  * (The user-facing side of this: https://systemd.io/BLOCK_DEVICE_LOCKING)
 471                  *
 472                  * There's a bit of a chicken and egg problem here for this however: inotify watching is
 473                  * supposed to be enabled via an option set via udev rules (OPTIONS+="watch"). If we skip the
 474                  * udev rules here however (as we just said we do), we would thus never see that specific
 475                  * udev rule, and thus never turn on inotify watching. But in order to catch up eventually
 476                  * and run them we we need the inotify watching: hence a classic chicken and egg problem.
 477                  *
 478                  * Our way out here: if we see the block device locked, unconditionally watch the device via
 479                  * inotify, regardless of any explicit request via OPTIONS+="watch". Thus, a device that is
 480                  * currently locked via the BSD file locks will be treated as if we ran a single udev rule
 481                  * only for it: the one that turns on inotify watching for it. If we eventually see the
 482                  * inotify IN_CLOSE_WRITE event, and then run the rules after all and we then realize that
 483                  * this wasn't actually requested (i.e. no OPTIONS+="watch" set) we'll simply turn off the
 484                  * watching again (see below). Effectively this means: inotify watching is now enabled either
 485                  * a) when the udev rules say so, or b) while the device is locked.
 486                  *
 487                  * Worst case scenario hence: in the (unlikely) case someone locked the device and we clash
 488                  * with that we might do inotify watching for a brief moment for a device where we actually
 489                  * weren't supposed to. But that shouldn't be too bad, in particular as BSD locks being taken
 490                  * on a block device is kinda an indication that the inotify logic is desired too, to some
 491                  * degree — they go hand-in-hand after all. */
 492
 493                 log_device_debug(dev, "Block device is currently locked, installing watch to wait until the lock is released.");
 494                 (void) udev_watch_begin(dev);
 495
 496                 /* Now the watch is installed, let's lock the device again, maybe in the meantime things changed */
 497                 r = worker_lock_block_device(dev, &fd_lock);
 498         }
 499         if (r < 0)
 500                 return r;
 501
 502         (void) worker_mark_block_device_read_only(dev);
 503
 504         /* apply rules, create node, symlinks */
 505         r = udev_event_execute_rules(udev_event, arg_event_timeout_usec, arg_timeout_signal, manager->properties, manager->rules);
 506         if (r < 0)
 507                 return r;
 508
 509         udev_event_execute_run(udev_event, arg_event_timeout_usec, arg_timeout_signal);
 510
 511         if (!manager->rtnl)
 512                 /* in case rtnl was initialized */
 513                 manager->rtnl = sd_netlink_ref(udev_event->rtnl);
 514
 515         /* apply/restore/end inotify watch */
 516         if (udev_event->inotify_watch) {
 517                 (void) udev_watch_begin(dev);
 518                 r = device_update_db(dev);
 519                 if (r < 0)
 520                         return log_device_debug_errno(dev, r, "Failed to update database under /run/udev/data/: %m");
 521         } else
 522                 (void) udev_watch_end(dev);
 523
 524         log_device_debug(dev, "Device (SEQNUM=%"PRIu64", ACTION=%s) processed",
 525                          seqnum, device_action_to_string(action));
 526
 527         return 0;
 528 }
 529
 530 static int worker_device_monitor_handler(sd_device_monitor *monitor, sd_device *dev, void *userdata) {
 531         Manager *manager = userdata;
 532         int r;
 533
 534         assert(dev);
 535         assert(manager);
 536
 537         r = worker_process_device(manager, dev);
 538         if (r == -EAGAIN)
 539                 /* if we couldn't acquire the flock(), then proceed quietly */
 540                 log_device_debug_errno(dev, r, "Device currently locked, not processing.");
 541         else {
 542                 if (r < 0)
 543                         log_device_warning_errno(dev, r, "Failed to process device, ignoring: %m");
 544
 545                 /* send processed event back to libudev listeners */
 546                 r = device_monitor_send_device(monitor, NULL, dev);
 547                 if (r < 0)
 548                         log_device_warning_errno(dev, r, "Failed to send device, ignoring: %m");
 549         }
 550
 551         /* send udevd the result of the event execution */
 552         r = worker_send_message(manager->worker_watch[WRITE_END]);
 553         if (r < 0)
 554                 log_device_warning_errno(dev, r, "Failed to send signal to main daemon, ignoring: %m");
 555
 556         return 1;
 557 }
 558
 559 static int worker_main(Manager *_manager, sd_device_monitor *monitor, sd_device *first_device) {
 560         _cleanup_(sd_device_unrefp) sd_device *dev = first_device;
 561         _cleanup_(manager_freep) Manager *manager = _manager;
 562         int r;
 563
 564         assert(manager);
 565         assert(monitor);
 566         assert(dev);
 567
 568         unsetenv("NOTIFY_SOCKET");
 569
 570         assert_se(sigprocmask_many(SIG_BLOCK, NULL, SIGTERM, -1) >= 0);
 571
 572         /* Reset OOM score, we only protect the main daemon. */
 573         r = set_oom_score_adjust(0);
 574         if (r < 0)
 575                 log_debug_errno(r, "Failed to reset OOM score, ignoring: %m");
 576
 577         /* Clear unnecessary data in Manager object.*/
 578         manager_clear_for_worker(manager);
 579
 580         r = sd_event_new(&manager->event);
 581         if (r < 0)
 582                 return log_error_errno(r, "Failed to allocate event loop: %m");
 583
 584         r = sd_event_add_signal(manager->event, NULL, SIGTERM, NULL, NULL);
 585         if (r < 0)
 586                 return log_error_errno(r, "Failed to set SIGTERM event: %m");
 587
 588         r = sd_device_monitor_attach_event(monitor, manager->event);
 589         if (r < 0)
 590                 return log_error_errno(r, "Failed to attach event loop to device monitor: %m");
 591
 592         r = sd_device_monitor_start(monitor, worker_device_monitor_handler, manager);
 593         if (r < 0)
 594                 return log_error_errno(r, "Failed to start device monitor: %m");
 595
 596         (void) sd_event_source_set_description(sd_device_monitor_get_event_source(monitor), "worker-device-monitor");
 597
 598         /* Process first device */
 599         (void) worker_device_monitor_handler(monitor, dev, manager);
 600
 601         r = sd_event_loop(manager->event);
 602         if (r < 0)
 603                 return log_error_errno(r, "Event loop failed: %m");
 604
 605         return 0;
 606 }
 607
 608 static int worker_spawn(Manager *manager, struct event *event) {
 609         _cleanup_(sd_device_monitor_unrefp) sd_device_monitor *worker_monitor = NULL;
 610         struct worker *worker;
 611         pid_t pid;
 612         int r;
 613
 614         /* listen for new events */
 615         r = device_monitor_new_full(&worker_monitor, MONITOR_GROUP_NONE, -1);
 616         if (r < 0)
 617                 return r;
 618
 619         /* allow the main daemon netlink address to send devices to the worker */
 620         r = device_monitor_allow_unicast_sender(worker_monitor, manager->monitor);
 621         if (r < 0)
 622                 return log_error_errno(r, "Worker: Failed to set unicast sender: %m");
 623
 624         r = device_monitor_enable_receiving(worker_monitor);
 625         if (r < 0)
 626                 return log_error_errno(r, "Worker: Failed to enable receiving of device: %m");
 627
 628         r = safe_fork(NULL, FORK_DEATHSIG, &pid);
 629         if (r < 0) {
 630                 event->state = EVENT_QUEUED;
 631                 return log_error_errno(r, "Failed to fork() worker: %m");
 632         }
 633         if (r == 0) {
 634                 /* Worker process */
 635                 r = worker_main(manager, worker_monitor, sd_device_ref(event->dev));
 636                 log_close();
 637                 _exit(r < 0 ? EXIT_FAILURE : EXIT_SUCCESS);
 638         }
 639
 640         r = worker_new(&worker, manager, worker_monitor, pid);
 641         if (r < 0)
 642                 return log_error_errno(r, "Failed to create worker object: %m");
 643
 644         worker_attach_event(worker, event);
 645
 646         log_device_debug(event->dev, "Worker ["PID_FMT"] is forked for processing SEQNUM=%"PRIu64".", pid, event->seqnum);
 647         return 0;
 648 }
 649
 650 static void event_run(Manager *manager, struct event *event) {
 651         static bool log_children_max_reached = true;
 652         struct worker *worker;
 653         int r;
 654
 655         assert(manager);
 656         assert(event);
 657
 658         if (DEBUG_LOGGING) {
 659                 DeviceAction action;
 660
 661                 r = device_get_action(event->dev, &action);
 662                 log_device_debug(event->dev, "Device (SEQNUM=%"PRIu64", ACTION=%s) ready for processing",
 663                                  event->seqnum, r >= 0 ? device_action_to_string(action) : "<unknown>");
 664         }
 665
 666         HASHMAP_FOREACH(worker, manager->workers) {
 667                 if (worker->state != WORKER_IDLE)
 668                         continue;
 669
 670                 r = device_monitor_send_device(manager->monitor, worker->monitor, event->dev);
 671                 if (r < 0) {
 672                         log_device_error_errno(event->dev, r, "Worker ["PID_FMT"] did not accept message, killing the worker: %m",
 673                                                worker->pid);
 674                         (void) kill(worker->pid, SIGKILL);
 675                         worker->state = WORKER_KILLED;
 676                         continue;
 677                 }
 678                 worker_attach_event(worker, event);
 679                 return;
 680         }
 681
 682         if (hashmap_size(manager->workers) >= arg_children_max) {
 683
 684                 /* Avoid spamming the debug logs if the limit is already reached and
 685                  * many events still need to be processed */
 686                 if (log_children_max_reached && arg_children_max > 1) {
 687                         log_debug("Maximum number (%u) of children reached.", hashmap_size(manager->workers));
 688                         log_children_max_reached = false;
 689                 }
 690                 return;
 691         }
 692
 693         /* Re-enable the debug message for the next batch of events */
 694         log_children_max_reached = true;
 695
 696         /* fork with up-to-date SELinux label database, so the child inherits the up-to-date db
 697            and, until the next SELinux policy changes, we safe further reloads in future children */
 698         mac_selinux_maybe_reload();
 699
 700         /* start new worker and pass initial device */
 701         worker_spawn(manager, event);
 702 }
 703
 704 static int event_queue_insert(Manager *manager, sd_device *dev) {
 705         _cleanup_(sd_device_unrefp) sd_device *clone = NULL;
 706         struct event *event;
 707         DeviceAction action;
 708         uint64_t seqnum;
 709         int r;
 710
 711         assert(manager);
 712         assert(dev);
 713
 714         /* only one process can add events to the queue */
 715         assert(manager->pid == getpid_cached());
 716
 717         /* We only accepts devices received by device monitor. */
 718         r = device_get_seqnum(dev, &seqnum);
 719         if (r < 0)
 720                 return r;
 721
 722         /* Refuse devices do not have ACTION property. */
 723         r = device_get_action(dev, &action);
 724         if (r < 0)
 725                 return r;
 726
 727         /* Save original device to restore the state on failures. */
 728         r = device_shallow_clone(dev, &clone);
 729         if (r < 0)
 730                 return r;
 731
 732         r = device_copy_properties(clone, dev);
 733         if (r < 0)
 734                 return r;
 735
 736         event = new(struct event, 1);
 737         if (!event)
 738                 return -ENOMEM;
 739
 740         *event = (struct event) {
 741                 .manager = manager,
 742                 .dev = sd_device_ref(dev),
 743                 .dev_kernel = TAKE_PTR(clone),
 744                 .seqnum = seqnum,
 745                 .state = EVENT_QUEUED,
 746         };
 747
 748         if (LIST_IS_EMPTY(manager->events)) {
 749                 r = touch("/run/udev/queue");
 750                 if (r < 0)
 751                         log_warning_errno(r, "Failed to touch /run/udev/queue: %m");
 752         }
 753
 754         LIST_APPEND(event, manager->events, event);
 755
 756         log_device_debug(dev, "Device (SEQNUM=%"PRIu64", ACTION=%s) is queued",
 757                          seqnum, device_action_to_string(action));
 758
 759         return 0;
 760 }
 761
 762 static void manager_kill_workers(Manager *manager) {
 763         struct worker *worker;
 764
 765         assert(manager);
 766
 767         HASHMAP_FOREACH(worker, manager->workers) {
 768                 if (worker->state == WORKER_KILLED)
 769                         continue;
 770
 771                 worker->state = WORKER_KILLED;
 772                 (void) kill(worker->pid, SIGTERM);
 773         }
 774 }
 775
 776 /* lookup event for identical, parent, child device */
 777 static int is_device_busy(Manager *manager, struct event *event) {
 778         const char *subsystem, *devpath, *devpath_old = NULL;
 779         dev_t devnum = makedev(0, 0);
 780         struct event *loop_event;
 781         size_t devpath_len;
 782         int r, ifindex = 0;
 783         bool is_block;
 784
 785         r = sd_device_get_subsystem(event->dev, &subsystem);
 786         if (r < 0)
 787                 return r;
 788
 789         is_block = streq(subsystem, "block");
 790
 791         r = sd_device_get_devpath(event->dev, &devpath);
 792         if (r < 0)
 793                 return r;
 794
 795         devpath_len = strlen(devpath);
 796
 797         r = sd_device_get_property_value(event->dev, "DEVPATH_OLD", &devpath_old);
 798         if (r < 0 && r != -ENOENT)
 799                 return r;
 800
 801         r = sd_device_get_devnum(event->dev, &devnum);
 802         if (r < 0 && r != -ENOENT)
 803                 return r;
 804
 805         r = sd_device_get_ifindex(event->dev, &ifindex);
 806         if (r < 0 && r != -ENOENT)
 807                 return r;
 808
 809         /* check if queue contains events we depend on */
 810         LIST_FOREACH(event, loop_event, manager->events) {
 811                 size_t loop_devpath_len, common;
 812                 const char *loop_devpath;
 813
 814                 /* we already found a later event, earlier cannot block us, no need to check again */
 815                 if (loop_event->seqnum < event->delaying_seqnum)
 816                         continue;
 817
 818                 /* event we checked earlier still exists, no need to check again */
 819                 if (loop_event->seqnum == event->delaying_seqnum)
 820                         return true;
 821
 822                 /* found ourself, no later event can block us */
 823                 if (loop_event->seqnum >= event->seqnum)
 824                         break;
 825
 826                 /* check major/minor */
 827                 if (major(devnum) != 0) {
 828                         const char *s;
 829                         dev_t d;
 830
 831                         if (sd_device_get_subsystem(loop_event->dev, &s) < 0)
 832                                 continue;
 833
 834                         if (sd_device_get_devnum(loop_event->dev, &d) >= 0 &&
 835                             devnum == d && is_block == streq(s, "block"))
 836                                 goto set_delaying_seqnum;
 837                 }
 838
 839                 /* check network device ifindex */
 840                 if (ifindex > 0) {
 841                         int i;
 842
 843                         if (sd_device_get_ifindex(loop_event->dev, &i) >= 0 &&
 844                             ifindex == i)
 845                                 goto set_delaying_seqnum;
 846                 }
 847
 848                 if (sd_device_get_devpath(loop_event->dev, &loop_devpath) < 0)
 849                         continue;
 850
 851                 /* check our old name */
 852                 if (devpath_old && streq(devpath_old, loop_devpath))
 853                         goto set_delaying_seqnum;
 854
 855                 loop_devpath_len = strlen(loop_devpath);
 856
 857                 /* compare devpath */
 858                 common = MIN(devpath_len, loop_devpath_len);
 859
 860                 /* one devpath is contained in the other? */
 861                 if (!strneq(devpath, loop_devpath, common))
 862                         continue;
 863
 864                 /* identical device event found */
 865                 if (devpath_len == loop_devpath_len)
 866                         goto set_delaying_seqnum;
 867
 868                 /* parent device event found */
 869                 if (devpath[common] == '/')
 870                         goto set_delaying_seqnum;
 871
 872                 /* child device event found */
 873                 if (loop_devpath[common] == '/')
 874                         goto set_delaying_seqnum;
 875         }
 876
 877         return false;
 878
 879 set_delaying_seqnum:
 880         log_device_debug(event->dev, "SEQNUM=%" PRIu64 " blocked by SEQNUM=%" PRIu64,
 881                          event->seqnum, loop_event->seqnum);
 882
 883         event->delaying_seqnum = loop_event->seqnum;
 884         return true;
 885 }
 886
 887 static void manager_exit(Manager *manager) {
 888         assert(manager);
 889
 890         manager->exit = true;
 891
 892         sd_notify(false,
 893                   "STOPPING=1\n"
 894                   "STATUS=Starting shutdown...");
 895
 896         /* close sources of new events and discard buffered events */
 897         manager->ctrl = udev_ctrl_unref(manager->ctrl);
 898
 899         manager->inotify_event = sd_event_source_unref(manager->inotify_event);
 900         manager->fd_inotify = safe_close(manager->fd_inotify);
 901
 902         manager->monitor = sd_device_monitor_unref(manager->monitor);
 903
 904         /* discard queued events and kill workers */
 905         event_queue_cleanup(manager, EVENT_QUEUED);
 906         manager_kill_workers(manager);
 907 }
 908
 909 /* reload requested, HUP signal received, rules changed, builtin changed */
 910 static void manager_reload(Manager *manager) {
 911
 912         assert(manager);
 913
 914         sd_notify(false,
 915                   "RELOADING=1\n"
 916                   "STATUS=Flushing configuration...");
 917
 918         manager_kill_workers(manager);
 919         manager->rules = udev_rules_free(manager->rules);
 920         udev_builtin_exit();
 921
 922         sd_notifyf(false,
 923                    "READY=1\n"
 924                    "STATUS=Processing with %u children at max", arg_children_max);
 925 }
 926
 927 static int on_kill_workers_event(sd_event_source *s, uint64_t usec, void *userdata) {
 928         Manager *manager = userdata;
 929
 930         assert(manager);
 931
 932         log_debug("Cleanup idle workers");
 933         manager_kill_workers(manager);
 934
 935         return 1;
 936 }
 937
 938 static void event_queue_start(Manager *manager) {
 939         struct event *event;
 940         usec_t usec;
 941         int r;
 942
 943         assert(manager);
 944
 945         if (LIST_IS_EMPTY(manager->events) ||
 946             manager->exit || manager->stop_exec_queue)
 947                 return;
 948
 949         assert_se(sd_event_now(manager->event, CLOCK_MONOTONIC, &usec) >= 0);
 950         /* check for changed config, every 3 seconds at most */
 951         if (manager->last_usec == 0 ||
 952             usec - manager->last_usec > 3 * USEC_PER_SEC) {
 953                 if (udev_rules_check_timestamp(manager->rules) ||
 954                     udev_builtin_validate())
 955                         manager_reload(manager);
 956
 957                 manager->last_usec = usec;
 958         }
 959
 960         r = event_source_disable(manager->kill_workers_event);
 961         if (r < 0)
 962                 log_warning_errno(r, "Failed to disable event source for cleaning up idle workers, ignoring: %m");
 963
 964         udev_builtin_init();
 965
 966         if (!manager->rules) {
 967                 r = udev_rules_load(&manager->rules, arg_resolve_name_timing);
 968                 if (r < 0) {
 969                         log_warning_errno(r, "Failed to read udev rules: %m");
 970                         return;
 971                 }
 972         }
 973
 974         LIST_FOREACH(event, event, manager->events) {
 975                 if (event->state != EVENT_QUEUED)
 976                         continue;
 977
 978                 /* do not start event if parent or child event is still running */
 979                 if (is_device_busy(manager, event) != 0)
 980                         continue;
 981
 982                 event_run(manager, event);
 983         }
 984 }
 985
 986 static void event_queue_cleanup(Manager *manager, enum event_state match_type) {
 987         struct event *event, *tmp;
 988
 989         LIST_FOREACH_SAFE(event, event, tmp, manager->events) {
 990                 if (match_type != EVENT_UNDEF && match_type != event->state)
 991                         continue;
 992
 993                 event_free(event);
 994         }
 995 }
 996
 997 static int on_worker(sd_event_source *s, int fd, uint32_t revents, void *userdata) {
 998         Manager *manager = userdata;
 999
1000         assert(manager);
1001
1002         for (;;) {
1003                 struct worker_message msg;
1004                 struct iovec iovec = {
1005                         .iov_base = &msg,
1006                         .iov_len = sizeof(msg),
1007                 };
1008                 CMSG_BUFFER_TYPE(CMSG_SPACE(sizeof(struct ucred))) control;
1009                 struct msghdr msghdr = {
1010                         .msg_iov = &iovec,
1011                         .msg_iovlen = 1,
1012                         .msg_control = &control,
1013                         .msg_controllen = sizeof(control),
1014                 };
1015                 ssize_t size;
1016                 struct ucred *ucred;
1017                 struct worker *worker;
1018
1019                 size = recvmsg_safe(fd, &msghdr, MSG_DONTWAIT);
1020                 if (size == -EINTR)
1021                         continue;
1022                 if (size == -EAGAIN)
1023                         /* nothing more to read */
1024                         break;
1025                 if (size < 0)
1026                         return log_error_errno(size, "Failed to receive message: %m");
1027
1028                 cmsg_close_all(&msghdr);
1029
1030                 if (size != sizeof(struct worker_message)) {
1031                         log_warning("Ignoring worker message with invalid size %zi bytes", size);
1032                         continue;
1033                 }
1034
1035                 ucred = CMSG_FIND_DATA(&msghdr, SOL_SOCKET, SCM_CREDENTIALS, struct ucred);
1036                 if (!ucred || ucred->pid <= 0) {
1037                         log_warning("Ignoring worker message without valid PID");
1038                         continue;
1039                 }
1040
1041                 /* lookup worker who sent the signal */
1042                 worker = hashmap_get(manager->workers, PID_TO_PTR(ucred->pid));
1043                 if (!worker) {
1044                         log_debug("Worker ["PID_FMT"] returned, but is no longer tracked", ucred->pid);
1045                         continue;
1046                 }
1047
1048                 if (worker->state != WORKER_KILLED)
1049                         worker->state = WORKER_IDLE;
1050
1051                 /* worker returned */
1052                 event_free(worker->event);
1053         }
1054
1055         /* we have free workers, try to schedule events */
1056         event_queue_start(manager);
1057
1058         return 1;
1059 }
1060
1061 static int on_uevent(sd_device_monitor *monitor, sd_device *dev, void *userdata) {
1062         Manager *manager = userdata;
1063         int r;
1064
1065         assert(manager);
1066
1067         device_ensure_usec_initialized(dev, NULL);
1068
1069         r = event_queue_insert(manager, dev);
1070         if (r < 0) {
1071                 log_device_error_errno(dev, r, "Failed to insert device into event queue: %m");
1072                 return 1;
1073         }
1074
1075         /* we have fresh events, try to schedule them */
1076         event_queue_start(manager);
1077
1078         return 1;
1079 }
1080
1081 /* receive the udevd message from userspace */
1082 static int on_ctrl_msg(struct udev_ctrl *uctrl, enum udev_ctrl_msg_type type, const union udev_ctrl_msg_value *value, void *userdata) {
1083         Manager *manager = userdata;
1084         int r;
1085
1086         assert(value);
1087         assert(manager);
1088
1089         switch (type) {
1090         case UDEV_CTRL_SET_LOG_LEVEL:
1091                 log_debug("Received udev control message (SET_LOG_LEVEL), setting log_level=%i", value->intval);
1092                 log_set_max_level_realm(LOG_REALM_UDEV, value->intval);
1093                 log_set_max_level_realm(LOG_REALM_SYSTEMD, value->intval);
1094                 manager_kill_workers(manager);
1095                 break;
1096         case UDEV_CTRL_STOP_EXEC_QUEUE:
1097                 log_debug("Received udev control message (STOP_EXEC_QUEUE)");
1098                 manager->stop_exec_queue = true;
1099                 break;
1100         case UDEV_CTRL_START_EXEC_QUEUE:
1101                 log_debug("Received udev control message (START_EXEC_QUEUE)");
1102                 manager->stop_exec_queue = false;
1103                 event_queue_start(manager);
1104                 break;
1105         case UDEV_CTRL_RELOAD:
1106                 log_debug("Received udev control message (RELOAD)");
1107                 manager_reload(manager);
1108                 break;
1109         case UDEV_CTRL_SET_ENV: {
1110                 _cleanup_free_ char *key = NULL, *val = NULL, *old_key = NULL, *old_val = NULL;
1111                 const char *eq;
1112
1113                 eq = strchr(value->buf, '=');
1114                 if (!eq) {
1115                         log_error("Invalid key format '%s'", value->buf);
1116                         return 1;
1117                 }
1118
1119                 key = strndup(value->buf, eq - value->buf);
1120                 if (!key) {
1121                         log_oom();
1122                         return 1;
1123                 }
1124
1125                 old_val = hashmap_remove2(manager->properties, key, (void **) &old_key);
1126
1127                 r = hashmap_ensure_allocated(&manager->properties, &string_hash_ops);
1128                 if (r < 0) {
1129                         log_oom();
1130                         return 1;
1131                 }
1132
1133                 eq++;
1134                 if (isempty(eq)) {
1135                         log_debug("Received udev control message (ENV), unsetting '%s'", key);
1136
1137                         r = hashmap_put(manager->properties, key, NULL);
1138                         if (r < 0) {
1139                                 log_oom();
1140                                 return 1;
1141                         }
1142                 } else {
1143                         val = strdup(eq);
1144                         if (!val) {
1145                                 log_oom();
1146                                 return 1;
1147                         }
1148
1149                         log_debug("Received udev control message (ENV), setting '%s=%s'", key, val);
1150
1151                         r = hashmap_put(manager->properties, key, val);
1152                         if (r < 0) {
1153                                 log_oom();
1154                                 return 1;
1155                         }
1156                 }
1157
1158                 key = val = NULL;
1159                 manager_kill_workers(manager);
1160                 break;
1161         }
1162         case UDEV_CTRL_SET_CHILDREN_MAX:
1163                 if (value->intval <= 0) {
1164                         log_debug("Received invalid udev control message (SET_MAX_CHILDREN, %i), ignoring.", value->intval);
1165                         return 0;
1166                 }
1167
1168                 log_debug("Received udev control message (SET_MAX_CHILDREN), setting children_max=%i", value->intval);
1169                 arg_children_max = value->intval;
1170
1171                 (void) sd_notifyf(false,
1172                                   "READY=1\n"
1173                                   "STATUS=Processing with %u children at max", arg_children_max);
1174                 break;
1175         case UDEV_CTRL_PING:
1176                 log_debug("Received udev control message (PING)");
1177                 break;
1178         case UDEV_CTRL_EXIT:
1179                 log_debug("Received udev control message (EXIT)");
1180                 manager_exit(manager);
1181                 break;
1182         default:
1183                 log_debug("Received unknown udev control message, ignoring");
1184         }
1185
1186         return 1;
1187 }
1188
1189 static int synthesize_change_one(sd_device *dev, const char *syspath) {
1190         const char *filename;
1191         int r;
1192
1193         filename = strjoina(syspath, "/uevent");
1194         log_device_debug(dev, "device is closed, synthesising 'change' on %s", syspath);
1195         r = write_string_file(filename, "change", WRITE_STRING_FILE_DISABLE_BUFFER);
1196         if (r < 0)
1197                 return log_device_debug_errno(dev, r, "Failed to write 'change' to %s: %m", filename);
1198         return 0;
1199 }
1200
1201 static int synthesize_change(sd_device *dev) {
1202         const char *subsystem, *sysname, *devname, *syspath, *devtype;
1203         int r;
1204
1205         r = sd_device_get_subsystem(dev, &subsystem);
1206         if (r < 0)
1207                 return r;
1208
1209         r = sd_device_get_sysname(dev, &sysname);
1210         if (r < 0)
1211                 return r;
1212
1213         r = sd_device_get_devname(dev, &devname);
1214         if (r < 0)
1215                 return r;
1216
1217         r = sd_device_get_syspath(dev, &syspath);
1218         if (r < 0)
1219                 return r;
1220
1221         r = sd_device_get_devtype(dev, &devtype);
1222         if (r < 0)
1223                 return r;
1224
1225         if (streq_ptr("block", subsystem) &&
1226             streq_ptr("disk", devtype) &&
1227             !startswith(sysname, "dm-")) {
1228                 _cleanup_(sd_device_enumerator_unrefp) sd_device_enumerator *e = NULL;
1229                 bool part_table_read = false, has_partitions = false;
1230                 sd_device *d;
1231                 int fd;
1232
1233                 /*
1234                  * Try to re-read the partition table. This only succeeds if
1235                  * none of the devices is busy. The kernel returns 0 if no
1236                  * partition table is found, and we will not get an event for
1237                  * the disk.
1238                  */
1239                 fd = open(devname, O_RDONLY|O_CLOEXEC|O_NOFOLLOW|O_NONBLOCK);
1240                 if (fd >= 0) {
1241                         r = flock(fd, LOCK_EX|LOCK_NB);
1242                         if (r >= 0)
1243                                 r = ioctl(fd, BLKRRPART, 0);
1244
1245                         close(fd);
1246                         if (r >= 0)
1247                                 part_table_read = true;
1248                 }
1249
1250                 /* search for partitions */
1251                 r = sd_device_enumerator_new(&e);
1252                 if (r < 0)
1253                         return r;
1254
1255                 r = sd_device_enumerator_allow_uninitialized(e);
1256                 if (r < 0)
1257                         return r;
1258
1259                 r = sd_device_enumerator_add_match_parent(e, dev);
1260                 if (r < 0)
1261                         return r;
1262
1263                 r = sd_device_enumerator_add_match_subsystem(e, "block", true);
1264                 if (r < 0)
1265                         return r;
1266
1267                 FOREACH_DEVICE(e, d) {
1268                         const char *t;
1269
1270                         if (sd_device_get_devtype(d, &t) < 0 ||
1271                             !streq("partition", t))
1272                                 continue;
1273
1274                         has_partitions = true;
1275                         break;
1276                 }
1277
1278                 /*
1279                  * We have partitions and re-read the table, the kernel already sent
1280                  * out a "change" event for the disk, and "remove/add" for all
1281                  * partitions.
1282                  */
1283                 if (part_table_read && has_partitions)
1284                         return 0;
1285
1286                 /*
1287                  * We have partitions but re-reading the partition table did not
1288                  * work, synthesize "change" for the disk and all partitions.
1289                  */
1290                 (void) synthesize_change_one(dev, syspath);
1291
1292                 FOREACH_DEVICE(e, d) {
1293                         const char *t, *n, *s;
1294
1295                         if (sd_device_get_devtype(d, &t) < 0 ||
1296                             !streq("partition", t))
1297                                 continue;
1298
1299                         if (sd_device_get_devname(d, &n) < 0 ||
1300                             sd_device_get_syspath(d, &s) < 0)
1301                                 continue;
1302
1303                         (void) synthesize_change_one(dev, s);
1304                 }
1305
1306         } else
1307                 (void) synthesize_change_one(dev, syspath);
1308
1309         return 0;
1310 }
1311
1312 static int on_inotify(sd_event_source *s, int fd, uint32_t revents, void *userdata) {
1313         Manager *manager = userdata;
1314         union inotify_event_buffer buffer;
1315         struct inotify_event *e;
1316         ssize_t l;
1317         int r;
1318
1319         assert(manager);
1320
1321         r = event_source_disable(manager->kill_workers_event);
1322         if (r < 0)
1323                 log_warning_errno(r, "Failed to disable event source for cleaning up idle workers, ignoring: %m");
1324
1325         l = read(fd, &buffer, sizeof(buffer));
1326         if (l < 0) {
1327                 if (IN_SET(errno, EAGAIN, EINTR))
1328                         return 1;
1329
1330                 return log_error_errno(errno, "Failed to read inotify fd: %m");
1331         }
1332
1333         FOREACH_INOTIFY_EVENT(e, buffer, l) {
1334                 _cleanup_(sd_device_unrefp) sd_device *dev = NULL;
1335                 const char *devnode;
1336
1337                 if (udev_watch_lookup(e->wd, &dev) <= 0)
1338                         continue;
1339
1340                 if (sd_device_get_devname(dev, &devnode) < 0)
1341                         continue;
1342
1343                 log_device_debug(dev, "Inotify event: %x for %s", e->mask, devnode);
1344                 if (e->mask & IN_CLOSE_WRITE)
1345                         synthesize_change(dev);
1346                 else if (e->mask & IN_IGNORED)
1347                         udev_watch_end(dev);
1348         }
1349
1350         return 1;
1351 }
1352
1353 static int on_sigterm(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
1354         Manager *manager = userdata;
1355
1356         assert(manager);
1357
1358         manager_exit(manager);
1359
1360         return 1;
1361 }
1362
1363 static int on_sighup(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
1364         Manager *manager = userdata;
1365
1366         assert(manager);
1367
1368         manager_reload(manager);
1369
1370         return 1;
1371 }
1372
1373 static int on_sigchld(sd_event_source *s, const struct signalfd_siginfo *si, void *userdata) {
1374         Manager *manager = userdata;
1375         int r;
1376
1377         assert(manager);
1378
1379         for (;;) {
1380                 pid_t pid;
1381                 int status;
1382                 struct worker *worker;
1383
1384                 pid = waitpid(-1, &status, WNOHANG);
1385                 if (pid <= 0)
1386                         break;
1387
1388                 worker = hashmap_get(manager->workers, PID_TO_PTR(pid));
1389                 if (!worker) {
1390                         log_warning("Worker ["PID_FMT"] is unknown, ignoring", pid);
1391                         continue;
1392                 }
1393
1394                 if (WIFEXITED(status)) {
1395                         if (WEXITSTATUS(status) == 0)
1396                                 log_debug("Worker ["PID_FMT"] exited", pid);
1397                         else
1398                                 log_warning("Worker ["PID_FMT"] exited with return code %i", pid, WEXITSTATUS(status));
1399                 } else if (WIFSIGNALED(status))
1400                         log_warning("Worker ["PID_FMT"] terminated by signal %i (%s)", pid, WTERMSIG(status), signal_to_string(WTERMSIG(status)));
1401                 else if (WIFSTOPPED(status)) {
1402                         log_info("Worker ["PID_FMT"] stopped", pid);
1403                         continue;
1404                 } else if (WIFCONTINUED(status)) {
1405                         log_info("Worker ["PID_FMT"] continued", pid);
1406                         continue;
1407                 } else
1408                         log_warning("Worker ["PID_FMT"] exit with status 0x%04x", pid, status);
1409
1410                 if ((!WIFEXITED(status) || WEXITSTATUS(status) != 0) && worker->event) {
1411                         log_device_error(worker->event->dev, "Worker ["PID_FMT"] failed", pid);
1412
1413                         /* delete state from disk */
1414                         device_delete_db(worker->event->dev);
1415                         device_tag_index(worker->event->dev, NULL, false);
1416
1417                         if (manager->monitor) {
1418                                 /* forward kernel event without amending it */
1419                                 r = device_monitor_send_device(manager->monitor, NULL, worker->event->dev_kernel);
1420                                 if (r < 0)
1421                                         log_device_error_errno(worker->event->dev_kernel, r, "Failed to send back device to kernel: %m");
1422                         }
1423                 }
1424
1425                 worker_free(worker);
1426         }
1427
1428         /* we can start new workers, try to schedule events */
1429         event_queue_start(manager);
1430
1431         /* Disable unnecessary cleanup event */
1432         if (hashmap_isempty(manager->workers)) {
1433                 r = event_source_disable(manager->kill_workers_event);
1434                 if (r < 0)
1435                         log_warning_errno(r, "Failed to disable event source for cleaning up idle workers, ignoring: %m");
1436         }
1437
1438         return 1;
1439 }
1440
1441 static int on_post(sd_event_source *s, void *userdata) {
1442         Manager *manager = userdata;
1443
1444         assert(manager);
1445
1446         if (!LIST_IS_EMPTY(manager->events))
1447                 return 1;
1448
1449         /* There are no pending events. Let's cleanup idle process. */
1450
1451         if (!hashmap_isempty(manager->workers)) {
1452                 /* There are idle workers */
1453                 (void) event_reset_time(manager->event, &manager->kill_workers_event, CLOCK_MONOTONIC,
1454                                         now(CLOCK_MONOTONIC) + 3 * USEC_PER_SEC, USEC_PER_SEC,
1455                                         on_kill_workers_event, manager, 0, "kill-workers-event", false);
1456                 return 1;
1457         }
1458
1459         /* There are no idle workers. */
1460
1461         if (manager->exit)
1462                 return sd_event_exit(manager->event, 0);
1463
1464         if (manager->cgroup)
1465                 /* cleanup possible left-over processes in our cgroup */
1466                 (void) cg_kill(SYSTEMD_CGROUP_CONTROLLER, manager->cgroup, SIGKILL, CGROUP_IGNORE_SELF, NULL, NULL, NULL);
1467
1468         return 1;
1469 }
1470
1471 static int listen_fds(int *ret_ctrl, int *ret_netlink) {
1472         int ctrl_fd = -1, netlink_fd = -1;
1473         int fd, n;
1474
1475         assert(ret_ctrl);
1476         assert(ret_netlink);
1477
1478         n = sd_listen_fds(true);
1479         if (n < 0)
1480                 return n;
1481
1482         for (fd = SD_LISTEN_FDS_START; fd < n + SD_LISTEN_FDS_START; fd++) {
1483                 if (sd_is_socket(fd, AF_LOCAL, SOCK_SEQPACKET, -1) > 0) {
1484                         if (ctrl_fd >= 0)
1485                                 return -EINVAL;
1486                         ctrl_fd = fd;
1487                         continue;
1488                 }
1489
1490                 if (sd_is_socket(fd, AF_NETLINK, SOCK_RAW, -1) > 0) {
1491                         if (netlink_fd >= 0)
1492                                 return -EINVAL;
1493                         netlink_fd = fd;
1494                         continue;
1495                 }
1496
1497                 return -EINVAL;
1498         }
1499
1500         *ret_ctrl = ctrl_fd;
1501         *ret_netlink = netlink_fd;
1502
1503         return 0;
1504 }
1505
1506 /*
1507  * read the kernel command line, in case we need to get into debug mode
1508  *   udev.log_level=<level>                    syslog priority
1509  *   udev.children_max=<number of workers>     events are fully serialized if set to 1
1510  *   udev.exec_delay=<number of seconds>       delay execution of every executed program
1511  *   udev.event_timeout=<number of seconds>    seconds to wait before terminating an event
1512  *   udev.blockdev_read_only<=bool>            mark all block devices read-only when they appear
1513  */
1514 static int parse_proc_cmdline_item(const char *key, const char *value, void *data) {
1515         int r;
1516
1517         assert(key);
1518
1519         if (proc_cmdline_key_streq(key, "udev.log_level") ||
1520             proc_cmdline_key_streq(key, "udev.log_priority")) { /* kept for backward compatibility */
1521
1522                 if (proc_cmdline_value_missing(key, value))
1523                         return 0;
1524
1525                 r = log_level_from_string(value);
1526                 if (r >= 0)
1527                         log_set_max_level(r);
1528
1529         } else if (proc_cmdline_key_streq(key, "udev.event_timeout")) {
1530
1531                 if (proc_cmdline_value_missing(key, value))
1532                         return 0;
1533
1534                 r = parse_sec(value, &arg_event_timeout_usec);
1535
1536         } else if (proc_cmdline_key_streq(key, "udev.children_max")) {
1537
1538                 if (proc_cmdline_value_missing(key, value))
1539                         return 0;
1540
1541                 r = safe_atou(value, &arg_children_max);
1542
1543         } else if (proc_cmdline_key_streq(key, "udev.exec_delay")) {
1544
1545                 if (proc_cmdline_value_missing(key, value))
1546                         return 0;
1547
1548                 r = parse_sec(value, &arg_exec_delay_usec);
1549
1550         } else if (proc_cmdline_key_streq(key, "udev.timeout_signal")) {
1551
1552                 if (proc_cmdline_value_missing(key, value))
1553                         return 0;
1554
1555                 r = signal_from_string(value);
1556                 if (r > 0)
1557                         arg_timeout_signal = r;
1558
1559         } else if (proc_cmdline_key_streq(key, "udev.blockdev_read_only")) {
1560
1561                 if (!value)
1562                         arg_blockdev_read_only = true;
1563                 else {
1564                         r = parse_boolean(value);
1565                         if (r < 0)
1566                                 log_warning_errno(r, "Failed to parse udev.blockdev-read-only argument, ignoring: %s", value);
1567                         else
1568                                 arg_blockdev_read_only = r;
1569                 }
1570
1571                 if (arg_blockdev_read_only)
1572                         log_notice("All physical block devices will be marked read-only.");
1573
1574                 return 0;
1575
1576         } else {
1577                 if (startswith(key, "udev."))
1578                         log_warning("Unknown udev kernel command line option \"%s\", ignoring.", key);
1579
1580                 return 0;
1581         }
1582
1583         if (r < 0)
1584                 log_warning_errno(r, "Failed to parse \"%s=%s\", ignoring: %m", key, value);
1585
1586         return 0;
1587 }
1588
1589 static int help(void) {
1590         _cleanup_free_ char *link = NULL;
1591         int r;
1592
1593         r = terminal_urlify_man("systemd-udevd.service", "8", &link);
1594         if (r < 0)
1595                 return log_oom();
1596
1597         printf("%s [OPTIONS...]\n\n"
1598                "Rule-based manager for device events and files.\n\n"
1599                "  -h --help                   Print this message\n"
1600                "  -V --version                Print version of the program\n"
1601                "  -d --daemon                 Detach and run in the background\n"
1602                "  -D --debug                  Enable debug output\n"
1603                "  -c --children-max=INT       Set maximum number of workers\n"
1604                "  -e --exec-delay=SECONDS     Seconds to wait before executing RUN=\n"
1605                "  -t --event-timeout=SECONDS  Seconds to wait before terminating an event\n"
1606                "  -N --resolve-names=early|late|never\n"
1607                "                              When to resolve users and groups\n"
1608                "\nSee the %s for details.\n"
1609                , program_invocation_short_name
1610                , link
1611         );
1612
1613         return 0;
1614 }
1615
1616 static int parse_argv(int argc, char *argv[]) {
1617         enum {
1618                 ARG_TIMEOUT_SIGNAL,
1619         };
1620
1621         static const struct option options[] = {
1622                 { "daemon",             no_argument,            NULL, 'd'                 },
1623                 { "debug",              no_argument,            NULL, 'D'                 },
1624                 { "children-max",       required_argument,      NULL, 'c'                 },
1625                 { "exec-delay",         required_argument,      NULL, 'e'                 },
1626                 { "event-timeout",      required_argument,      NULL, 't'                 },
1627                 { "resolve-names",      required_argument,      NULL, 'N'                 },
1628                 { "help",               no_argument,            NULL, 'h'                 },
1629                 { "version",            no_argument,            NULL, 'V'                 },
1630                 { "timeout-signal",     required_argument,      NULL,  ARG_TIMEOUT_SIGNAL },
1631                 {}
1632         };
1633
1634         int c, r;
1635
1636         assert(argc >= 0);
1637         assert(argv);
1638
1639         while ((c = getopt_long(argc, argv, "c:de:Dt:N:hV", options, NULL)) >= 0) {
1640                 switch (c) {
1641
1642                 case 'd':
1643                         arg_daemonize = true;
1644                         break;
1645                 case 'c':
1646                         r = safe_atou(optarg, &arg_children_max);
1647                         if (r < 0)
1648                                 log_warning_errno(r, "Failed to parse --children-max= value '%s', ignoring: %m", optarg);
1649                         break;
1650                 case 'e':
1651                         r = parse_sec(optarg, &arg_exec_delay_usec);
1652                         if (r < 0)
1653                                 log_warning_errno(r, "Failed to parse --exec-delay= value '%s', ignoring: %m", optarg);
1654                         break;
1655                 case ARG_TIMEOUT_SIGNAL:
1656                         r = signal_from_string(optarg);
1657                         if (r <= 0)
1658                                 log_warning_errno(r, "Failed to parse --timeout-signal= value '%s', ignoring: %m", optarg);
1659                         else
1660                                 arg_timeout_signal = r;
1661
1662                         break;
1663                 case 't':
1664                         r = parse_sec(optarg, &arg_event_timeout_usec);
1665                         if (r < 0)
1666                                 log_warning_errno(r, "Failed to parse --event-timeout= value '%s', ignoring: %m", optarg);
1667                         break;
1668                 case 'D':
1669                         arg_debug = true;
1670                         break;
1671                 case 'N': {
1672                         ResolveNameTiming t;
1673
1674                         t = resolve_name_timing_from_string(optarg);
1675                         if (t < 0)
1676                                 log_warning("Invalid --resolve-names= value '%s', ignoring.", optarg);
1677                         else
1678                                 arg_resolve_name_timing = t;
1679                         break;
1680                 }
1681                 case 'h':
1682                         return help();
1683                 case 'V':
1684                         printf("%s\n", GIT_VERSION);
1685                         return 0;
1686                 case '?':
1687                         return -EINVAL;
1688                 default:
1689                         assert_not_reached("Unhandled option");
1690
1691                 }
1692         }
1693
1694         return 1;
1695 }
1696
1697 static int manager_new(Manager **ret, int fd_ctrl, int fd_uevent, const char *cgroup) {
1698         _cleanup_(manager_freep) Manager *manager = NULL;
1699         int r;
1700
1701         assert(ret);
1702
1703         manager = new(Manager, 1);
1704         if (!manager)
1705                 return log_oom();
1706
1707         *manager = (Manager) {
1708                 .fd_inotify = -1,
1709                 .worker_watch = { -1, -1 },
1710                 .cgroup = cgroup,
1711         };
1712
1713         r = udev_ctrl_new_from_fd(&manager->ctrl, fd_ctrl);
1714         if (r < 0)
1715                 return log_error_errno(r, "Failed to initialize udev control socket: %m");
1716
1717         r = udev_ctrl_enable_receiving(manager->ctrl);
1718         if (r < 0)
1719                 return log_error_errno(r, "Failed to bind udev control socket: %m");
1720
1721         r = device_monitor_new_full(&manager->monitor, MONITOR_GROUP_KERNEL, fd_uevent);
1722         if (r < 0)
1723                 return log_error_errno(r, "Failed to initialize device monitor: %m");
1724
1725         /* Bump receiver buffer, but only if we are not called via socket activation, as in that
1726          * case systemd sets the receive buffer size for us, and the value in the .socket unit
1727          * should take full effect. */
1728         if (fd_uevent < 0) {
1729                 r = sd_device_monitor_set_receive_buffer_size(manager->monitor, 128 * 1024 * 1024);
1730                 if (r < 0)
1731                         log_warning_errno(r, "Failed to set receive buffer size for device monitor, ignoring: %m");
1732         }
1733
1734         r = device_monitor_enable_receiving(manager->monitor);
1735         if (r < 0)
1736                 return log_error_errno(r, "Failed to bind netlink socket: %m");
1737
1738         *ret = TAKE_PTR(manager);
1739
1740         return 0;
1741 }
1742
1743 static int main_loop(Manager *manager) {
1744         int fd_worker, r;
1745
1746         manager->pid = getpid_cached();
1747
1748         /* unnamed socket from workers to the main daemon */
1749         r = socketpair(AF_LOCAL, SOCK_DGRAM|SOCK_CLOEXEC, 0, manager->worker_watch);
1750         if (r < 0)
1751                 return log_error_errno(errno, "Failed to create socketpair for communicating with workers: %m");
1752
1753         fd_worker = manager->worker_watch[READ_END];
1754
1755         r = setsockopt_int(fd_worker, SOL_SOCKET, SO_PASSCRED, true);
1756         if (r < 0)
1757                 return log_error_errno(r, "Failed to enable SO_PASSCRED: %m");
1758
1759         r = udev_watch_init();
1760         if (r < 0)
1761                 return log_error_errno(r, "Failed to create inotify descriptor: %m");
1762         manager->fd_inotify = r;
1763
1764         udev_watch_restore();
1765
1766         /* block and listen to all signals on signalfd */
1767         assert_se(sigprocmask_many(SIG_BLOCK, NULL, SIGTERM, SIGINT, SIGHUP, SIGCHLD, -1) >= 0);
1768
1769         r = sd_event_default(&manager->event);
1770         if (r < 0)
1771                 return log_error_errno(r, "Failed to allocate event loop: %m");
1772
1773         r = sd_event_add_signal(manager->event, NULL, SIGINT, on_sigterm, manager);
1774         if (r < 0)
1775                 return log_error_errno(r, "Failed to create SIGINT event source: %m");
1776
1777         r = sd_event_add_signal(manager->event, NULL, SIGTERM, on_sigterm, manager);
1778         if (r < 0)
1779                 return log_error_errno(r, "Failed to create SIGTERM event source: %m");
1780
1781         r = sd_event_add_signal(manager->event, NULL, SIGHUP, on_sighup, manager);
1782         if (r < 0)
1783                 return log_error_errno(r, "Failed to create SIGHUP event source: %m");
1784
1785         r = sd_event_add_signal(manager->event, NULL, SIGCHLD, on_sigchld, manager);
1786         if (r < 0)
1787                 return log_error_errno(r, "Failed to create SIGCHLD event source: %m");
1788
1789         r = sd_event_set_watchdog(manager->event, true);
1790         if (r < 0)
1791                 return log_error_errno(r, "Failed to create watchdog event source: %m");
1792
1793         r = udev_ctrl_attach_event(manager->ctrl, manager->event);
1794         if (r < 0)
1795                 return log_error_errno(r, "Failed to attach event to udev control: %m");
1796
1797         r = udev_ctrl_start(manager->ctrl, on_ctrl_msg, manager);
1798         if (r < 0)
1799                 return log_error_errno(r, "Failed to start device monitor: %m");
1800
1801         /* This needs to be after the inotify and uevent handling, to make sure
1802          * that the ping is send back after fully processing the pending uevents
1803          * (including the synthetic ones we may create due to inotify events).
1804          */
1805         r = sd_event_source_set_priority(udev_ctrl_get_event_source(manager->ctrl), SD_EVENT_PRIORITY_IDLE);
1806         if (r < 0)
1807                 return log_error_errno(r, "Failed to set IDLE event priority for udev control event source: %m");
1808
1809         r = sd_event_add_io(manager->event, &manager->inotify_event, manager->fd_inotify, EPOLLIN, on_inotify, manager);
1810         if (r < 0)
1811                 return log_error_errno(r, "Failed to create inotify event source: %m");
1812
1813         r = sd_device_monitor_attach_event(manager->monitor, manager->event);
1814         if (r < 0)
1815                 return log_error_errno(r, "Failed to attach event to device monitor: %m");
1816
1817         r = sd_device_monitor_start(manager->monitor, on_uevent, manager);
1818         if (r < 0)
1819                 return log_error_errno(r, "Failed to start device monitor: %m");
1820
1821         (void) sd_event_source_set_description(sd_device_monitor_get_event_source(manager->monitor), "device-monitor");
1822
1823         r = sd_event_add_io(manager->event, NULL, fd_worker, EPOLLIN, on_worker, manager);
1824         if (r < 0)
1825                 return log_error_errno(r, "Failed to create worker event source: %m");
1826
1827         r = sd_event_add_post(manager->event, NULL, on_post, manager);
1828         if (r < 0)
1829                 return log_error_errno(r, "Failed to create post event source: %m");
1830
1831         udev_builtin_init();
1832
1833         r = udev_rules_load(&manager->rules, arg_resolve_name_timing);
1834         if (!manager->rules)
1835                 return log_error_errno(r, "Failed to read udev rules: %m");
1836
1837         r = udev_rules_apply_static_dev_perms(manager->rules);
1838         if (r < 0)
1839                 log_error_errno(r, "Failed to apply permissions on static device nodes: %m");
1840
1841         (void) sd_notifyf(false,
1842                           "READY=1\n"
1843                           "STATUS=Processing with %u children at max", arg_children_max);
1844
1845         r = sd_event_loop(manager->event);
1846         if (r < 0)
1847                 log_error_errno(r, "Event loop failed: %m");
1848
1849         sd_notify(false,
1850                   "STOPPING=1\n"
1851                   "STATUS=Shutting down...");
1852         return r;
1853 }
1854
1855 int run_udevd(int argc, char *argv[]) {
1856         _cleanup_free_ char *cgroup = NULL;
1857         _cleanup_(manager_freep) Manager *manager = NULL;
1858         int fd_ctrl = -1, fd_uevent = -1;
1859         int r;
1860
1861         log_set_target(LOG_TARGET_AUTO);
1862         log_open();
1863         udev_parse_config_full(&arg_children_max, &arg_exec_delay_usec, &arg_event_timeout_usec, &arg_resolve_name_timing, &arg_timeout_signal);
1864         log_parse_environment();
1865         log_open(); /* Done again to update after reading configuration. */
1866
1867         r = parse_argv(argc, argv);
1868         if (r <= 0)
1869                 return r;
1870
1871         r = proc_cmdline_parse(parse_proc_cmdline_item, NULL, PROC_CMDLINE_STRIP_RD_PREFIX);
1872         if (r < 0)
1873                 log_warning_errno(r, "Failed to parse kernel command line, ignoring: %m");
1874
1875         if (arg_debug) {
1876                 log_set_target(LOG_TARGET_CONSOLE);
1877                 log_set_max_level(LOG_DEBUG);
1878         }
1879
1880         log_set_max_level_realm(LOG_REALM_SYSTEMD, log_get_max_level());
1881
1882         r = must_be_root();
1883         if (r < 0)
1884                 return r;
1885
1886         if (arg_children_max == 0) {
1887                 unsigned long cpu_limit, mem_limit, cpu_count = 1;
1888
1889                 r = cpus_in_affinity_mask();
1890                 if (r < 0)
1891                         log_warning_errno(r, "Failed to determine number of local CPUs, ignoring: %m");
1892                 else
1893                         cpu_count = r;
1894
1895                 cpu_limit = cpu_count * 2 + 16;
1896                 mem_limit = MAX(physical_memory() / (128UL*1024*1024), 10U);
1897
1898                 arg_children_max = MIN(cpu_limit, mem_limit);
1899                 arg_children_max = MIN(WORKER_NUM_MAX, arg_children_max);
1900
1901                 log_debug("Set children_max to %u", arg_children_max);
1902         }
1903
1904         /* set umask before creating any file/directory */
1905         umask(022);
1906
1907         r = mac_selinux_init();
1908         if (r < 0)
1909                 return r;
1910
1911         r = mkdir_errno_wrapper("/run/udev", 0755);
1912         if (r < 0 && r != -EEXIST)
1913                 return log_error_errno(r, "Failed to create /run/udev: %m");
1914
1915         if (getppid() == 1 && sd_booted() > 0) {
1916                 /* Get our own cgroup, we regularly kill everything udev has left behind.
1917                  * We only do this on systemd systems, and only if we are directly spawned
1918                  * by PID1. Otherwise we are not guaranteed to have a dedicated cgroup. */
1919                 r = cg_pid_get_path(SYSTEMD_CGROUP_CONTROLLER, 0, &cgroup);
1920                 if (r < 0) {
1921                         if (IN_SET(r, -ENOENT, -ENOMEDIUM))
1922                                 log_debug_errno(r, "Dedicated cgroup not found: %m");
1923                         else
1924                                 log_warning_errno(r, "Failed to get cgroup: %m");
1925                 }
1926         }
1927
1928         r = listen_fds(&fd_ctrl, &fd_uevent);
1929         if (r < 0)
1930                 return log_error_errno(r, "Failed to listen on fds: %m");
1931
1932         r = manager_new(&manager, fd_ctrl, fd_uevent, cgroup);
1933         if (r < 0)
1934                 return log_error_errno(r, "Failed to create manager: %m");
1935
1936         if (arg_daemonize) {
1937                 pid_t pid;
1938
1939                 log_info("Starting version " GIT_VERSION);
1940
1941                 /* connect /dev/null to stdin, stdout, stderr */
1942                 if (log_get_max_level() < LOG_DEBUG) {
1943                         r = make_null_stdio();
1944                         if (r < 0)
1945                                 log_warning_errno(r, "Failed to redirect standard streams to /dev/null: %m");
1946                 }
1947
1948                 pid = fork();
1949                 if (pid < 0)
1950                         return log_error_errno(errno, "Failed to fork daemon: %m");
1951                 if (pid > 0)
1952                         /* parent */
1953                         return 0;
1954
1955                 /* child */
1956                 (void) setsid();
1957         }
1958
1959         return main_loop(manager);
1960 }