src/core/manager.c

   1 /***
   2   This file is part of systemd.
   3
   4   Copyright 2010 Lennart Poettering
   5
   6   systemd is free software; you can redistribute it and/or modify it
   7   under the terms of the GNU Lesser General Public License as published by
   8   the Free Software Foundation; either version 2.1 of the License, or
   9   (at your option) any later version.
  10
  11   systemd is distributed in the hope that it will be useful, but
  12   WITHOUT ANY WARRANTY; without even the implied warranty of
  13   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  14   Lesser General Public License for more details.
  15
  16   You should have received a copy of the GNU Lesser General Public License
  17   along with systemd; If not, see <http://www.gnu.org/licenses/>.
  18 ***/
  19
  20 #include <errno.h>
  21 #include <fcntl.h>
  22 #include <linux/kd.h>
  23 #include <signal.h>
  24 #include <string.h>
  25 #include <sys/epoll.h>
  26 #include <sys/inotify.h>
  27 #include <sys/ioctl.h>
  28 #include <sys/reboot.h>
  29 #include <sys/timerfd.h>
  30 #include <sys/wait.h>
  31 #include <unistd.h>
  32
  33 #ifdef HAVE_AUDIT
  34 #include <libaudit.h>
  35 #endif
  36
  37 #include "sd-daemon.h"
  38 #include "sd-messages.h"
  39
  40 #include "alloc-util.h"
  41 #include "audit-fd.h"
  42 #include "boot-timestamps.h"
  43 #include "bus-common-errors.h"
  44 #include "bus-error.h"
  45 #include "bus-kernel.h"
  46 #include "bus-util.h"
  47 #include "clean-ipc.h"
  48 #include "dbus-job.h"
  49 #include "dbus-manager.h"
  50 #include "dbus-unit.h"
  51 #include "dbus.h"
  52 #include "dirent-util.h"
  53 #include "env-util.h"
  54 #include "escape.h"
  55 #include "exit-status.h"
  56 #include "fd-util.h"
  57 #include "fileio.h"
  58 #include "fs-util.h"
  59 #include "hashmap.h"
  60 #include "io-util.h"
  61 #include "locale-setup.h"
  62 #include "log.h"
  63 #include "macro.h"
  64 #include "manager.h"
  65 #include "missing.h"
  66 #include "mkdir.h"
  67 #include "parse-util.h"
  68 #include "path-lookup.h"
  69 #include "path-util.h"
  70 #include "process-util.h"
  71 #include "ratelimit.h"
  72 #include "rm-rf.h"
  73 #include "signal-util.h"
  74 #include "special.h"
  75 #include "stat-util.h"
  76 #include "string-table.h"
  77 #include "string-util.h"
  78 #include "strv.h"
  79 #include "terminal-util.h"
  80 #include "time-util.h"
  81 #include "transaction.h"
  82 #include "umask-util.h"
  83 #include "unit-name.h"
  84 #include "user-util.h"
  85 #include "util.h"
  86 #include "virt.h"
  87 #include "watchdog.h"
  88
  89 #define NOTIFY_RCVBUF_SIZE (8*1024*1024)
  90 #define CGROUPS_AGENT_RCVBUF_SIZE (8*1024*1024)
  91
  92 /* Initial delay and the interval for printing status messages about running jobs */
  93 #define JOBS_IN_PROGRESS_WAIT_USEC (5*USEC_PER_SEC)
  94 #define JOBS_IN_PROGRESS_PERIOD_USEC (USEC_PER_SEC / 3)
  95 #define JOBS_IN_PROGRESS_PERIOD_DIVISOR 3
  96
  97 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
  98 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
  99 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
 100 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
 101 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
 102 static int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata);
 103 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata);
 104 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata);
 105 static int manager_run_generators(Manager *m);
 106
 107 static void manager_watch_jobs_in_progress(Manager *m) {
 108         usec_t next;
 109         int r;
 110
 111         assert(m);
 112
 113         /* We do not want to show the cylon animation if the user
 114          * needs to confirm service executions otherwise confirmation
 115          * messages will be screwed by the cylon animation. */
 116         if (!manager_is_confirm_spawn_disabled(m))
 117                 return;
 118
 119         if (m->jobs_in_progress_event_source)
 120                 return;
 121
 122         next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC;
 123         r = sd_event_add_time(
 124                         m->event,
 125                         &m->jobs_in_progress_event_source,
 126                         CLOCK_MONOTONIC,
 127                         next, 0,
 128                         manager_dispatch_jobs_in_progress, m);
 129         if (r < 0)
 130                 return;
 131
 132         (void) sd_event_source_set_description(m->jobs_in_progress_event_source, "manager-jobs-in-progress");
 133 }
 134
 135 #define CYLON_BUFFER_EXTRA (2*(sizeof(ANSI_RED)-1) + sizeof(ANSI_HIGHLIGHT_RED)-1 + 2*(sizeof(ANSI_NORMAL)-1))
 136
 137 static void draw_cylon(char buffer[], size_t buflen, unsigned width, unsigned pos) {
 138         char *p = buffer;
 139
 140         assert(buflen >= CYLON_BUFFER_EXTRA + width + 1);
 141         assert(pos <= width+1); /* 0 or width+1 mean that the center light is behind the corner */
 142
 143         if (pos > 1) {
 144                 if (pos > 2)
 145                         p = mempset(p, ' ', pos-2);
 146                 if (log_get_show_color())
 147                         p = stpcpy(p, ANSI_RED);
 148                 *p++ = '*';
 149         }
 150
 151         if (pos > 0 && pos <= width) {
 152                 if (log_get_show_color())
 153                         p = stpcpy(p, ANSI_HIGHLIGHT_RED);
 154                 *p++ = '*';
 155         }
 156
 157         if (log_get_show_color())
 158                 p = stpcpy(p, ANSI_NORMAL);
 159
 160         if (pos < width) {
 161                 if (log_get_show_color())
 162                         p = stpcpy(p, ANSI_RED);
 163                 *p++ = '*';
 164                 if (pos < width-1)
 165                         p = mempset(p, ' ', width-1-pos);
 166                 if (log_get_show_color())
 167                         strcpy(p, ANSI_NORMAL);
 168         }
 169 }
 170
 171 void manager_flip_auto_status(Manager *m, bool enable) {
 172         assert(m);
 173
 174         if (enable) {
 175                 if (m->show_status == SHOW_STATUS_AUTO)
 176                         manager_set_show_status(m, SHOW_STATUS_TEMPORARY);
 177         } else {
 178                 if (m->show_status == SHOW_STATUS_TEMPORARY)
 179                         manager_set_show_status(m, SHOW_STATUS_AUTO);
 180         }
 181 }
 182
 183 static void manager_print_jobs_in_progress(Manager *m) {
 184         _cleanup_free_ char *job_of_n = NULL;
 185         Iterator i;
 186         Job *j;
 187         unsigned counter = 0, print_nr;
 188         char cylon[6 + CYLON_BUFFER_EXTRA + 1];
 189         unsigned cylon_pos;
 190         char time[FORMAT_TIMESPAN_MAX], limit[FORMAT_TIMESPAN_MAX] = "no limit";
 191         uint64_t x;
 192
 193         assert(m);
 194         assert(m->n_running_jobs > 0);
 195
 196         manager_flip_auto_status(m, true);
 197
 198         print_nr = (m->jobs_in_progress_iteration / JOBS_IN_PROGRESS_PERIOD_DIVISOR) % m->n_running_jobs;
 199
 200         HASHMAP_FOREACH(j, m->jobs, i)
 201                 if (j->state == JOB_RUNNING && counter++ == print_nr)
 202                         break;
 203
 204         /* m->n_running_jobs must be consistent with the contents of m->jobs,
 205          * so the above loop must have succeeded in finding j. */
 206         assert(counter == print_nr + 1);
 207         assert(j);
 208
 209         cylon_pos = m->jobs_in_progress_iteration % 14;
 210         if (cylon_pos >= 8)
 211                 cylon_pos = 14 - cylon_pos;
 212         draw_cylon(cylon, sizeof(cylon), 6, cylon_pos);
 213
 214         m->jobs_in_progress_iteration++;
 215
 216         if (m->n_running_jobs > 1) {
 217                 if (asprintf(&job_of_n, "(%u of %u) ", counter, m->n_running_jobs) < 0)
 218                         job_of_n = NULL;
 219         }
 220
 221         format_timespan(time, sizeof(time), now(CLOCK_MONOTONIC) - j->begin_usec, 1*USEC_PER_SEC);
 222         if (job_get_timeout(j, &x) > 0)
 223                 format_timespan(limit, sizeof(limit), x - j->begin_usec, 1*USEC_PER_SEC);
 224
 225         manager_status_printf(m, STATUS_TYPE_EPHEMERAL, cylon,
 226                               "%sA %s job is running for %s (%s / %s)",
 227                               strempty(job_of_n),
 228                               job_type_to_string(j->type),
 229                               unit_description(j->unit),
 230                               time, limit);
 231 }
 232
 233 static int have_ask_password(void) {
 234         _cleanup_closedir_ DIR *dir;
 235         struct dirent *de;
 236
 237         dir = opendir("/run/systemd/ask-password");
 238         if (!dir) {
 239                 if (errno == ENOENT)
 240                         return false;
 241                 else
 242                         return -errno;
 243         }
 244
 245         FOREACH_DIRENT_ALL(de, dir, return -errno) {
 246                 if (startswith(de->d_name, "ask."))
 247                         return true;
 248         }
 249         return false;
 250 }
 251
 252 static int manager_dispatch_ask_password_fd(sd_event_source *source,
 253                                             int fd, uint32_t revents, void *userdata) {
 254         Manager *m = userdata;
 255
 256         assert(m);
 257
 258         flush_fd(fd);
 259
 260         m->have_ask_password = have_ask_password();
 261         if (m->have_ask_password < 0)
 262                 /* Log error but continue. Negative have_ask_password
 263                  * is treated as unknown status. */
 264                 log_error_errno(m->have_ask_password, "Failed to list /run/systemd/ask-password: %m");
 265
 266         return 0;
 267 }
 268
 269 static void manager_close_ask_password(Manager *m) {
 270         assert(m);
 271
 272         m->ask_password_event_source = sd_event_source_unref(m->ask_password_event_source);
 273         m->ask_password_inotify_fd = safe_close(m->ask_password_inotify_fd);
 274         m->have_ask_password = -EINVAL;
 275 }
 276
 277 static int manager_check_ask_password(Manager *m) {
 278         int r;
 279
 280         assert(m);
 281
 282         if (!m->ask_password_event_source) {
 283                 assert(m->ask_password_inotify_fd < 0);
 284
 285                 mkdir_p_label("/run/systemd/ask-password", 0755);
 286
 287                 m->ask_password_inotify_fd = inotify_init1(IN_NONBLOCK|IN_CLOEXEC);
 288                 if (m->ask_password_inotify_fd < 0)
 289                         return log_error_errno(errno, "inotify_init1() failed: %m");
 290
 291                 if (inotify_add_watch(m->ask_password_inotify_fd, "/run/systemd/ask-password", IN_CREATE|IN_DELETE|IN_MOVE) < 0) {
 292                         log_error_errno(errno, "Failed to add watch on /run/systemd/ask-password: %m");
 293                         manager_close_ask_password(m);
 294                         return -errno;
 295                 }
 296
 297                 r = sd_event_add_io(m->event, &m->ask_password_event_source,
 298                                     m->ask_password_inotify_fd, EPOLLIN,
 299                                     manager_dispatch_ask_password_fd, m);
 300                 if (r < 0) {
 301                         log_error_errno(errno, "Failed to add event source for /run/systemd/ask-password: %m");
 302                         manager_close_ask_password(m);
 303                         return -errno;
 304                 }
 305
 306                 (void) sd_event_source_set_description(m->ask_password_event_source, "manager-ask-password");
 307
 308                 /* Queries might have been added meanwhile... */
 309                 manager_dispatch_ask_password_fd(m->ask_password_event_source,
 310                                                  m->ask_password_inotify_fd, EPOLLIN, m);
 311         }
 312
 313         return m->have_ask_password;
 314 }
 315
 316 static int manager_watch_idle_pipe(Manager *m) {
 317         int r;
 318
 319         assert(m);
 320
 321         if (m->idle_pipe_event_source)
 322                 return 0;
 323
 324         if (m->idle_pipe[2] < 0)
 325                 return 0;
 326
 327         r = sd_event_add_io(m->event, &m->idle_pipe_event_source, m->idle_pipe[2], EPOLLIN, manager_dispatch_idle_pipe_fd, m);
 328         if (r < 0)
 329                 return log_error_errno(r, "Failed to watch idle pipe: %m");
 330
 331         (void) sd_event_source_set_description(m->idle_pipe_event_source, "manager-idle-pipe");
 332
 333         return 0;
 334 }
 335
 336 static void manager_close_idle_pipe(Manager *m) {
 337         assert(m);
 338
 339         m->idle_pipe_event_source = sd_event_source_unref(m->idle_pipe_event_source);
 340
 341         safe_close_pair(m->idle_pipe);
 342         safe_close_pair(m->idle_pipe + 2);
 343 }
 344
 345 static int manager_setup_time_change(Manager *m) {
 346         int r;
 347
 348         /* We only care for the cancellation event, hence we set the
 349          * timeout to the latest possible value. */
 350         struct itimerspec its = {
 351                 .it_value.tv_sec = TIME_T_MAX,
 352         };
 353
 354         assert(m);
 355         assert_cc(sizeof(time_t) == sizeof(TIME_T_MAX));
 356
 357         if (m->test_run)
 358                 return 0;
 359
 360         /* Uses TFD_TIMER_CANCEL_ON_SET to get notifications whenever
 361          * CLOCK_REALTIME makes a jump relative to CLOCK_MONOTONIC */
 362
 363         m->time_change_fd = timerfd_create(CLOCK_REALTIME, TFD_NONBLOCK|TFD_CLOEXEC);
 364         if (m->time_change_fd < 0)
 365                 return log_error_errno(errno, "Failed to create timerfd: %m");
 366
 367         if (timerfd_settime(m->time_change_fd, TFD_TIMER_ABSTIME|TFD_TIMER_CANCEL_ON_SET, &its, NULL) < 0) {
 368                 log_debug_errno(errno, "Failed to set up TFD_TIMER_CANCEL_ON_SET, ignoring: %m");
 369                 m->time_change_fd = safe_close(m->time_change_fd);
 370                 return 0;
 371         }
 372
 373         r = sd_event_add_io(m->event, &m->time_change_event_source, m->time_change_fd, EPOLLIN, manager_dispatch_time_change_fd, m);
 374         if (r < 0)
 375                 return log_error_errno(r, "Failed to create time change event source: %m");
 376
 377         (void) sd_event_source_set_description(m->time_change_event_source, "manager-time-change");
 378
 379         log_debug("Set up TFD_TIMER_CANCEL_ON_SET timerfd.");
 380
 381         return 0;
 382 }
 383
 384 static int enable_special_signals(Manager *m) {
 385         _cleanup_close_ int fd = -1;
 386
 387         assert(m);
 388
 389         if (m->test_run)
 390                 return 0;
 391
 392         /* Enable that we get SIGINT on control-alt-del. In containers
 393          * this will fail with EPERM (older) or EINVAL (newer), so
 394          * ignore that. */
 395         if (reboot(RB_DISABLE_CAD) < 0 && errno != EPERM && errno != EINVAL)
 396                 log_warning_errno(errno, "Failed to enable ctrl-alt-del handling: %m");
 397
 398         fd = open_terminal("/dev/tty0", O_RDWR|O_NOCTTY|O_CLOEXEC);
 399         if (fd < 0) {
 400                 /* Support systems without virtual console */
 401                 if (fd != -ENOENT)
 402                         log_warning_errno(errno, "Failed to open /dev/tty0: %m");
 403         } else {
 404                 /* Enable that we get SIGWINCH on kbrequest */
 405                 if (ioctl(fd, KDSIGACCEPT, SIGWINCH) < 0)
 406                         log_warning_errno(errno, "Failed to enable kbrequest handling: %m");
 407         }
 408
 409         return 0;
 410 }
 411
 412 static int manager_setup_signals(Manager *m) {
 413         struct sigaction sa = {
 414                 .sa_handler = SIG_DFL,
 415                 .sa_flags = SA_NOCLDSTOP|SA_RESTART,
 416         };
 417         sigset_t mask;
 418         int r;
 419
 420         assert(m);
 421
 422         assert_se(sigaction(SIGCHLD, &sa, NULL) == 0);
 423
 424         /* We make liberal use of realtime signals here. On
 425          * Linux/glibc we have 30 of them (with the exception of Linux
 426          * on hppa, see below), between SIGRTMIN+0 ... SIGRTMIN+30
 427          * (aka SIGRTMAX). */
 428
 429         assert_se(sigemptyset(&mask) == 0);
 430         sigset_add_many(&mask,
 431                         SIGCHLD,     /* Child died */
 432                         SIGTERM,     /* Reexecute daemon */
 433                         SIGHUP,      /* Reload configuration */
 434                         SIGUSR1,     /* systemd/upstart: reconnect to D-Bus */
 435                         SIGUSR2,     /* systemd: dump status */
 436                         SIGINT,      /* Kernel sends us this on control-alt-del */
 437                         SIGWINCH,    /* Kernel sends us this on kbrequest (alt-arrowup) */
 438                         SIGPWR,      /* Some kernel drivers and upsd send us this on power failure */
 439
 440                         SIGRTMIN+0,  /* systemd: start default.target */
 441                         SIGRTMIN+1,  /* systemd: isolate rescue.target */
 442                         SIGRTMIN+2,  /* systemd: isolate emergency.target */
 443                         SIGRTMIN+3,  /* systemd: start halt.target */
 444                         SIGRTMIN+4,  /* systemd: start poweroff.target */
 445                         SIGRTMIN+5,  /* systemd: start reboot.target */
 446                         SIGRTMIN+6,  /* systemd: start kexec.target */
 447
 448                         /* ... space for more special targets ... */
 449
 450                         SIGRTMIN+13, /* systemd: Immediate halt */
 451                         SIGRTMIN+14, /* systemd: Immediate poweroff */
 452                         SIGRTMIN+15, /* systemd: Immediate reboot */
 453                         SIGRTMIN+16, /* systemd: Immediate kexec */
 454
 455                         /* ... space for more immediate system state changes ... */
 456
 457                         SIGRTMIN+20, /* systemd: enable status messages */
 458                         SIGRTMIN+21, /* systemd: disable status messages */
 459                         SIGRTMIN+22, /* systemd: set log level to LOG_DEBUG */
 460                         SIGRTMIN+23, /* systemd: set log level to LOG_INFO */
 461                         SIGRTMIN+24, /* systemd: Immediate exit (--user only) */
 462
 463                         /* .. one free signal here ... */
 464
 465 #if !defined(__hppa64__) && !defined(__hppa__)
 466                         /* Apparently Linux on hppa has fewer RT
 467                          * signals (SIGRTMAX is SIGRTMIN+25 there),
 468                          * hence let's not try to make use of them
 469                          * here. Since these commands are accessible
 470                          * by different means and only really a safety
 471                          * net, the missing functionality on hppa
 472                          * shouldn't matter. */
 473
 474                         SIGRTMIN+26, /* systemd: set log target to journal-or-kmsg */
 475                         SIGRTMIN+27, /* systemd: set log target to console */
 476                         SIGRTMIN+28, /* systemd: set log target to kmsg */
 477                         SIGRTMIN+29, /* systemd: set log target to syslog-or-kmsg (obsolete) */
 478
 479                         /* ... one free signal here SIGRTMIN+30 ... */
 480 #endif
 481                         -1);
 482         assert_se(sigprocmask(SIG_SETMASK, &mask, NULL) == 0);
 483
 484         m->signal_fd = signalfd(-1, &mask, SFD_NONBLOCK|SFD_CLOEXEC);
 485         if (m->signal_fd < 0)
 486                 return -errno;
 487
 488         r = sd_event_add_io(m->event, &m->signal_event_source, m->signal_fd, EPOLLIN, manager_dispatch_signal_fd, m);
 489         if (r < 0)
 490                 return r;
 491
 492         (void) sd_event_source_set_description(m->signal_event_source, "manager-signal");
 493
 494         /* Process signals a bit earlier than the rest of things, but later than notify_fd processing, so that the
 495          * notify processing can still figure out to which process/service a message belongs, before we reap the
 496          * process. Also, process this before handling cgroup notifications, so that we always collect child exit
 497          * status information before detecting that there's no process in a cgroup. */
 498         r = sd_event_source_set_priority(m->signal_event_source, SD_EVENT_PRIORITY_NORMAL-6);
 499         if (r < 0)
 500                 return r;
 501
 502         if (MANAGER_IS_SYSTEM(m))
 503                 return enable_special_signals(m);
 504
 505         return 0;
 506 }
 507
 508 static void manager_clean_environment(Manager *m) {
 509         assert(m);
 510
 511         /* Let's remove some environment variables that we
 512          * need ourselves to communicate with our clients */
 513         strv_env_unset_many(
 514                         m->environment,
 515                         "NOTIFY_SOCKET",
 516                         "MAINPID",
 517                         "MANAGERPID",
 518                         "LISTEN_PID",
 519                         "LISTEN_FDS",
 520                         "LISTEN_FDNAMES",
 521                         "WATCHDOG_PID",
 522                         "WATCHDOG_USEC",
 523                         "INVOCATION_ID",
 524                         NULL);
 525 }
 526
 527 static int manager_default_environment(Manager *m) {
 528         assert(m);
 529
 530         if (MANAGER_IS_SYSTEM(m)) {
 531                 /* The system manager always starts with a clean
 532                  * environment for its children. It does not import
 533                  * the kernel or the parents exported variables.
 534                  *
 535                  * The initial passed environ is untouched to keep
 536                  * /proc/self/environ valid; it is used for tagging
 537                  * the init process inside containers. */
 538                 m->environment = strv_new("PATH=" DEFAULT_PATH,
 539                                           NULL);
 540
 541                 /* Import locale variables LC_*= from configuration */
 542                 locale_setup(&m->environment);
 543         } else {
 544                 /* The user manager passes its own environment
 545                  * along to its children. */
 546                 m->environment = strv_copy(environ);
 547         }
 548
 549         if (!m->environment)
 550                 return -ENOMEM;
 551
 552         manager_clean_environment(m);
 553         strv_sort(m->environment);
 554
 555         return 0;
 556 }
 557
 558 int manager_new(UnitFileScope scope, bool test_run, Manager **_m) {
 559         Manager *m;
 560         int r;
 561
 562         assert(_m);
 563         assert(IN_SET(scope, UNIT_FILE_SYSTEM, UNIT_FILE_USER));
 564
 565         m = new0(Manager, 1);
 566         if (!m)
 567                 return -ENOMEM;
 568
 569         m->unit_file_scope = scope;
 570         m->exit_code = _MANAGER_EXIT_CODE_INVALID;
 571         m->default_timer_accuracy_usec = USEC_PER_MINUTE;
 572         m->default_tasks_accounting = true;
 573         m->default_tasks_max = UINT64_MAX;
 574
 575 #ifdef ENABLE_EFI
 576         if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0)
 577                 boot_timestamps(&m->userspace_timestamp, &m->firmware_timestamp, &m->loader_timestamp);
 578 #endif
 579
 580         /* Prepare log fields we can use for structured logging */
 581         if (MANAGER_IS_SYSTEM(m)) {
 582                 m->unit_log_field = "UNIT=";
 583                 m->unit_log_format_string = "UNIT=%s";
 584
 585                 m->invocation_log_field = "INVOCATION_ID=";
 586                 m->invocation_log_format_string = "INVOCATION_ID=" SD_ID128_FORMAT_STR;
 587         } else {
 588                 m->unit_log_field = "USER_UNIT=";
 589                 m->unit_log_format_string = "USER_UNIT=%s";
 590
 591                 m->invocation_log_field = "USER_INVOCATION_ID=";
 592                 m->invocation_log_format_string = "USER_INVOCATION_ID=" SD_ID128_FORMAT_STR;
 593         }
 594
 595         m->idle_pipe[0] = m->idle_pipe[1] = m->idle_pipe[2] = m->idle_pipe[3] = -1;
 596
 597         m->pin_cgroupfs_fd = m->notify_fd = m->cgroups_agent_fd = m->signal_fd = m->time_change_fd =
 598                 m->dev_autofs_fd = m->private_listen_fd = m->cgroup_inotify_fd =
 599                 m->ask_password_inotify_fd = -1;
 600
 601         m->user_lookup_fds[0] = m->user_lookup_fds[1] = -1;
 602
 603         m->current_job_id = 1; /* start as id #1, so that we can leave #0 around as "null-like" value */
 604
 605         m->have_ask_password = -EINVAL; /* we don't know */
 606         m->first_boot = -1;
 607
 608         m->test_run = test_run;
 609
 610         /* Reboot immediately if the user hits C-A-D more often than 7x per 2s */
 611         RATELIMIT_INIT(m->ctrl_alt_del_ratelimit, 2 * USEC_PER_SEC, 7);
 612
 613         r = manager_default_environment(m);
 614         if (r < 0)
 615                 goto fail;
 616
 617         r = hashmap_ensure_allocated(&m->units, &string_hash_ops);
 618         if (r < 0)
 619                 goto fail;
 620
 621         r = hashmap_ensure_allocated(&m->jobs, NULL);
 622         if (r < 0)
 623                 goto fail;
 624
 625         r = hashmap_ensure_allocated(&m->cgroup_unit, &string_hash_ops);
 626         if (r < 0)
 627                 goto fail;
 628
 629         r = hashmap_ensure_allocated(&m->watch_bus, &string_hash_ops);
 630         if (r < 0)
 631                 goto fail;
 632
 633         r = sd_event_default(&m->event);
 634         if (r < 0)
 635                 goto fail;
 636
 637         r = sd_event_add_defer(m->event, &m->run_queue_event_source, manager_dispatch_run_queue, m);
 638         if (r < 0)
 639                 goto fail;
 640
 641         r = sd_event_source_set_priority(m->run_queue_event_source, SD_EVENT_PRIORITY_IDLE);
 642         if (r < 0)
 643                 goto fail;
 644
 645         r = sd_event_source_set_enabled(m->run_queue_event_source, SD_EVENT_OFF);
 646         if (r < 0)
 647                 goto fail;
 648
 649         (void) sd_event_source_set_description(m->run_queue_event_source, "manager-run-queue");
 650
 651         r = manager_setup_signals(m);
 652         if (r < 0)
 653                 goto fail;
 654
 655         r = manager_setup_cgroup(m);
 656         if (r < 0)
 657                 goto fail;
 658
 659         r = manager_setup_time_change(m);
 660         if (r < 0)
 661                 goto fail;
 662
 663         m->udev = udev_new();
 664         if (!m->udev) {
 665                 r = -ENOMEM;
 666                 goto fail;
 667         }
 668
 669         /* Note that we do not set up the notify fd here. We do that after deserialization,
 670          * since they might have gotten serialized across the reexec. */
 671
 672         m->taint_usr = dir_is_empty("/usr") > 0;
 673
 674         *_m = m;
 675         return 0;
 676
 677 fail:
 678         manager_free(m);
 679         return r;
 680 }
 681
 682 static int manager_setup_notify(Manager *m) {
 683         int r;
 684
 685         if (m->test_run)
 686                 return 0;
 687
 688         if (m->notify_fd < 0) {
 689                 _cleanup_close_ int fd = -1;
 690                 union sockaddr_union sa = {
 691                         .sa.sa_family = AF_UNIX,
 692                 };
 693                 static const int one = 1;
 694                 const char *e;
 695
 696                 /* First free all secondary fields */
 697                 m->notify_socket = mfree(m->notify_socket);
 698                 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
 699
 700                 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
 701                 if (fd < 0)
 702                         return log_error_errno(errno, "Failed to allocate notification socket: %m");
 703
 704                 fd_inc_rcvbuf(fd, NOTIFY_RCVBUF_SIZE);
 705
 706                 e = manager_get_runtime_prefix(m);
 707                 if (!e) {
 708                         log_error("Failed to determine runtime prefix.");
 709                         return -EINVAL;
 710                 }
 711
 712                 m->notify_socket = strappend(e, "/systemd/notify");
 713                 if (!m->notify_socket)
 714                         return log_oom();
 715
 716                 (void) mkdir_parents_label(m->notify_socket, 0755);
 717                 (void) unlink(m->notify_socket);
 718
 719                 strncpy(sa.un.sun_path, m->notify_socket, sizeof(sa.un.sun_path)-1);
 720                 r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
 721                 if (r < 0)
 722                         return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
 723
 724                 r = setsockopt(fd, SOL_SOCKET, SO_PASSCRED, &one, sizeof(one));
 725                 if (r < 0)
 726                         return log_error_errno(errno, "SO_PASSCRED failed: %m");
 727
 728                 m->notify_fd = fd;
 729                 fd = -1;
 730
 731                 log_debug("Using notification socket %s", m->notify_socket);
 732         }
 733
 734         if (!m->notify_event_source) {
 735                 r = sd_event_add_io(m->event, &m->notify_event_source, m->notify_fd, EPOLLIN, manager_dispatch_notify_fd, m);
 736                 if (r < 0)
 737                         return log_error_errno(r, "Failed to allocate notify event source: %m");
 738
 739                 /* Process notification messages a bit earlier than SIGCHLD, so that we can still identify to which
 740                  * service an exit message belongs. */
 741                 r = sd_event_source_set_priority(m->notify_event_source, SD_EVENT_PRIORITY_NORMAL-7);
 742                 if (r < 0)
 743                         return log_error_errno(r, "Failed to set priority of notify event source: %m");
 744
 745                 (void) sd_event_source_set_description(m->notify_event_source, "manager-notify");
 746         }
 747
 748         return 0;
 749 }
 750
 751 static int manager_setup_cgroups_agent(Manager *m) {
 752
 753         static const union sockaddr_union sa = {
 754                 .un.sun_family = AF_UNIX,
 755                 .un.sun_path = "/run/systemd/cgroups-agent",
 756         };
 757         int r;
 758
 759         /* This creates a listening socket we receive cgroups agent messages on. We do not use D-Bus for delivering
 760          * these messages from the cgroups agent binary to PID 1, as the cgroups agent binary is very short-living, and
 761          * each instance of it needs a new D-Bus connection. Since D-Bus connections are SOCK_STREAM/AF_UNIX, on
 762          * overloaded systems the backlog of the D-Bus socket becomes relevant, as not more than the configured number
 763          * of D-Bus connections may be queued until the kernel will start dropping further incoming connections,
 764          * possibly resulting in lost cgroups agent messages. To avoid this, we'll use a private SOCK_DGRAM/AF_UNIX
 765          * socket, where no backlog is relevant as communication may take place without an actual connect() cycle, and
 766          * we thus won't lose messages.
 767          *
 768          * Note that PID 1 will forward the agent message to system bus, so that the user systemd instance may listen
 769          * to it. The system instance hence listens on this special socket, but the user instances listen on the system
 770          * bus for these messages. */
 771
 772         if (m->test_run)
 773                 return 0;
 774
 775         if (!MANAGER_IS_SYSTEM(m))
 776                 return 0;
 777
 778         if (cg_unified(SYSTEMD_CGROUP_CONTROLLER) > 0) /* We don't need this anymore on the unified hierarchy */
 779                 return 0;
 780
 781         if (m->cgroups_agent_fd < 0) {
 782                 _cleanup_close_ int fd = -1;
 783
 784                 /* First free all secondary fields */
 785                 m->cgroups_agent_event_source = sd_event_source_unref(m->cgroups_agent_event_source);
 786
 787                 fd = socket(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
 788                 if (fd < 0)
 789                         return log_error_errno(errno, "Failed to allocate cgroups agent socket: %m");
 790
 791                 fd_inc_rcvbuf(fd, CGROUPS_AGENT_RCVBUF_SIZE);
 792
 793                 (void) unlink(sa.un.sun_path);
 794
 795                 /* Only allow root to connect to this socket */
 796                 RUN_WITH_UMASK(0077)
 797                         r = bind(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un));
 798                 if (r < 0)
 799                         return log_error_errno(errno, "bind(%s) failed: %m", sa.un.sun_path);
 800
 801                 m->cgroups_agent_fd = fd;
 802                 fd = -1;
 803         }
 804
 805         if (!m->cgroups_agent_event_source) {
 806                 r = sd_event_add_io(m->event, &m->cgroups_agent_event_source, m->cgroups_agent_fd, EPOLLIN, manager_dispatch_cgroups_agent_fd, m);
 807                 if (r < 0)
 808                         return log_error_errno(r, "Failed to allocate cgroups agent event source: %m");
 809
 810                 /* Process cgroups notifications early, but after having processed service notification messages or
 811                  * SIGCHLD signals, so that a cgroup running empty is always just the last safety net of notification,
 812                  * and we collected the metadata the notification and SIGCHLD stuff offers first. Also see handling of
 813                  * cgroup inotify for the unified cgroup stuff. */
 814                 r = sd_event_source_set_priority(m->cgroups_agent_event_source, SD_EVENT_PRIORITY_NORMAL-5);
 815                 if (r < 0)
 816                         return log_error_errno(r, "Failed to set priority of cgroups agent event source: %m");
 817
 818                 (void) sd_event_source_set_description(m->cgroups_agent_event_source, "manager-cgroups-agent");
 819         }
 820
 821         return 0;
 822 }
 823
 824 static int manager_setup_user_lookup_fd(Manager *m) {
 825         int r;
 826
 827         assert(m);
 828
 829         /* Set up the socket pair used for passing UID/GID resolution results from forked off processes to PID
 830          * 1. Background: we can't do name lookups (NSS) from PID 1, since it might involve IPC and thus activation,
 831          * and we might hence deadlock on ourselves. Hence we do all user/group lookups asynchronously from the forked
 832          * off processes right before executing the binaries to start. In order to be able to clean up any IPC objects
 833          * created by a unit (see RemoveIPC=) we need to know in PID 1 the used UID/GID of the executed processes,
 834          * hence we establish this communication channel so that forked off processes can pass their UID/GID
 835          * information back to PID 1. The forked off processes send their resolved UID/GID to PID 1 in a simple
 836          * datagram, along with their unit name, so that we can share one communication socket pair among all units for
 837          * this purpose.
 838          *
 839          * You might wonder why we need a communication channel for this that is independent of the usual notification
 840          * socket scheme (i.e. $NOTIFY_SOCKET). The primary difference is about trust: data sent via the $NOTIFY_SOCKET
 841          * channel is only accepted if it originates from the right unit and if reception was enabled for it. The user
 842          * lookup socket OTOH is only accessible by PID 1 and its children until they exec(), and always available.
 843          *
 844          * Note that this function is called under two circumstances: when we first initialize (in which case we
 845          * allocate both the socket pair and the event source to listen on it), and when we deserialize after a reload
 846          * (in which case the socket pair already exists but we still need to allocate the event source for it). */
 847
 848         if (m->user_lookup_fds[0] < 0) {
 849
 850                 /* Free all secondary fields */
 851                 safe_close_pair(m->user_lookup_fds);
 852                 m->user_lookup_event_source = sd_event_source_unref(m->user_lookup_event_source);
 853
 854                 if (socketpair(AF_UNIX, SOCK_DGRAM|SOCK_CLOEXEC, 0, m->user_lookup_fds) < 0)
 855                         return log_error_errno(errno, "Failed to allocate user lookup socket: %m");
 856
 857                 (void) fd_inc_rcvbuf(m->user_lookup_fds[0], NOTIFY_RCVBUF_SIZE);
 858         }
 859
 860         if (!m->user_lookup_event_source) {
 861                 r = sd_event_add_io(m->event, &m->user_lookup_event_source, m->user_lookup_fds[0], EPOLLIN, manager_dispatch_user_lookup_fd, m);
 862                 if (r < 0)
 863                         return log_error_errno(errno, "Failed to allocate user lookup event source: %m");
 864
 865                 /* Process even earlier than the notify event source, so that we always know first about valid UID/GID
 866                  * resolutions */
 867                 r = sd_event_source_set_priority(m->user_lookup_event_source, SD_EVENT_PRIORITY_NORMAL-8);
 868                 if (r < 0)
 869                         return log_error_errno(errno, "Failed to set priority ot user lookup event source: %m");
 870
 871                 (void) sd_event_source_set_description(m->user_lookup_event_source, "user-lookup");
 872         }
 873
 874         return 0;
 875 }
 876
 877 static int manager_connect_bus(Manager *m, bool reexecuting) {
 878         bool try_bus_connect;
 879
 880         assert(m);
 881
 882         if (m->test_run)
 883                 return 0;
 884
 885         try_bus_connect =
 886                 reexecuting ||
 887                 (MANAGER_IS_USER(m) && getenv("DBUS_SESSION_BUS_ADDRESS"));
 888
 889         /* Try to connect to the buses, if possible. */
 890         return bus_init(m, try_bus_connect);
 891 }
 892
 893 static unsigned manager_dispatch_cleanup_queue(Manager *m) {
 894         Unit *u;
 895         unsigned n = 0;
 896
 897         assert(m);
 898
 899         while ((u = m->cleanup_queue)) {
 900                 assert(u->in_cleanup_queue);
 901
 902                 unit_free(u);
 903                 n++;
 904         }
 905
 906         return n;
 907 }
 908
 909 enum {
 910         GC_OFFSET_IN_PATH,  /* This one is on the path we were traveling */
 911         GC_OFFSET_UNSURE,   /* No clue */
 912         GC_OFFSET_GOOD,     /* We still need this unit */
 913         GC_OFFSET_BAD,      /* We don't need this unit anymore */
 914         _GC_OFFSET_MAX
 915 };
 916
 917 static void unit_gc_mark_good(Unit *u, unsigned gc_marker) {
 918         Iterator i;
 919         Unit *other;
 920
 921         u->gc_marker = gc_marker + GC_OFFSET_GOOD;
 922
 923         /* Recursively mark referenced units as GOOD as well */
 924         SET_FOREACH(other, u->dependencies[UNIT_REFERENCES], i)
 925                 if (other->gc_marker == gc_marker + GC_OFFSET_UNSURE)
 926                         unit_gc_mark_good(other, gc_marker);
 927 }
 928
 929 static void unit_gc_sweep(Unit *u, unsigned gc_marker) {
 930         Iterator i;
 931         Unit *other;
 932         bool is_bad;
 933
 934         assert(u);
 935
 936         if (u->gc_marker == gc_marker + GC_OFFSET_GOOD ||
 937             u->gc_marker == gc_marker + GC_OFFSET_BAD ||
 938             u->gc_marker == gc_marker + GC_OFFSET_UNSURE ||
 939             u->gc_marker == gc_marker + GC_OFFSET_IN_PATH)
 940                 return;
 941
 942         if (u->in_cleanup_queue)
 943                 goto bad;
 944
 945         if (unit_check_gc(u))
 946                 goto good;
 947
 948         u->gc_marker = gc_marker + GC_OFFSET_IN_PATH;
 949
 950         is_bad = true;
 951
 952         SET_FOREACH(other, u->dependencies[UNIT_REFERENCED_BY], i) {
 953                 unit_gc_sweep(other, gc_marker);
 954
 955                 if (other->gc_marker == gc_marker + GC_OFFSET_GOOD)
 956                         goto good;
 957
 958                 if (other->gc_marker != gc_marker + GC_OFFSET_BAD)
 959                         is_bad = false;
 960         }
 961
 962         if (is_bad)
 963                 goto bad;
 964
 965         /* We were unable to find anything out about this entry, so
 966          * let's investigate it later */
 967         u->gc_marker = gc_marker + GC_OFFSET_UNSURE;
 968         unit_add_to_gc_queue(u);
 969         return;
 970
 971 bad:
 972         /* We definitely know that this one is not useful anymore, so
 973          * let's mark it for deletion */
 974         u->gc_marker = gc_marker + GC_OFFSET_BAD;
 975         unit_add_to_cleanup_queue(u);
 976         return;
 977
 978 good:
 979         unit_gc_mark_good(u, gc_marker);
 980 }
 981
 982 static unsigned manager_dispatch_gc_unit_queue(Manager *m) {
 983         unsigned n = 0, gc_marker;
 984         Unit *u;
 985
 986         assert(m);
 987
 988         /* log_debug("Running GC..."); */
 989
 990         m->gc_marker += _GC_OFFSET_MAX;
 991         if (m->gc_marker + _GC_OFFSET_MAX <= _GC_OFFSET_MAX)
 992                 m->gc_marker = 1;
 993
 994         gc_marker = m->gc_marker;
 995
 996         while ((u = m->gc_unit_queue)) {
 997                 assert(u->in_gc_queue);
 998
 999                 unit_gc_sweep(u, gc_marker);
1000
1001                 LIST_REMOVE(gc_queue, m->gc_unit_queue, u);
1002                 u->in_gc_queue = false;
1003
1004                 n++;
1005
1006                 if (u->gc_marker == gc_marker + GC_OFFSET_BAD ||
1007                     u->gc_marker == gc_marker + GC_OFFSET_UNSURE) {
1008                         if (u->id)
1009                                 log_unit_debug(u, "Collecting.");
1010                         u->gc_marker = gc_marker + GC_OFFSET_BAD;
1011                         unit_add_to_cleanup_queue(u);
1012                 }
1013         }
1014
1015         return n;
1016 }
1017
1018 static unsigned manager_dispatch_gc_job_queue(Manager *m) {
1019         unsigned n = 0;
1020         Job *j;
1021
1022         assert(m);
1023
1024         while ((j = m->gc_job_queue)) {
1025                 assert(j->in_gc_queue);
1026
1027                 LIST_REMOVE(gc_queue, m->gc_job_queue, j);
1028                 j->in_gc_queue = false;
1029
1030                 n++;
1031
1032                 if (job_check_gc(j))
1033                         continue;
1034
1035                 log_unit_debug(j->unit, "Collecting job.");
1036                 (void) job_finish_and_invalidate(j, JOB_COLLECTED, false, false);
1037         }
1038
1039         return n;
1040 }
1041
1042 static void manager_clear_jobs_and_units(Manager *m) {
1043         Unit *u;
1044
1045         assert(m);
1046
1047         while ((u = hashmap_first(m->units)))
1048                 unit_free(u);
1049
1050         manager_dispatch_cleanup_queue(m);
1051
1052         assert(!m->load_queue);
1053         assert(!m->run_queue);
1054         assert(!m->dbus_unit_queue);
1055         assert(!m->dbus_job_queue);
1056         assert(!m->cleanup_queue);
1057         assert(!m->gc_unit_queue);
1058         assert(!m->gc_job_queue);
1059
1060         assert(hashmap_isempty(m->jobs));
1061         assert(hashmap_isempty(m->units));
1062
1063         m->n_on_console = 0;
1064         m->n_running_jobs = 0;
1065 }
1066
1067 Manager* manager_free(Manager *m) {
1068         UnitType c;
1069         int i;
1070
1071         if (!m)
1072                 return NULL;
1073
1074         manager_clear_jobs_and_units(m);
1075
1076         for (c = 0; c < _UNIT_TYPE_MAX; c++)
1077                 if (unit_vtable[c]->shutdown)
1078                         unit_vtable[c]->shutdown(m);
1079
1080         /* If we reexecute ourselves, we keep the root cgroup
1081          * around */
1082         manager_shutdown_cgroup(m, m->exit_code != MANAGER_REEXECUTE);
1083
1084         lookup_paths_flush_generator(&m->lookup_paths);
1085
1086         bus_done(m);
1087
1088         dynamic_user_vacuum(m, false);
1089         hashmap_free(m->dynamic_users);
1090
1091         hashmap_free(m->units);
1092         hashmap_free(m->units_by_invocation_id);
1093         hashmap_free(m->jobs);
1094         hashmap_free(m->watch_pids1);
1095         hashmap_free(m->watch_pids2);
1096         hashmap_free(m->watch_bus);
1097
1098         set_free(m->startup_units);
1099         set_free(m->failed_units);
1100
1101         sd_event_source_unref(m->signal_event_source);
1102         sd_event_source_unref(m->notify_event_source);
1103         sd_event_source_unref(m->cgroups_agent_event_source);
1104         sd_event_source_unref(m->time_change_event_source);
1105         sd_event_source_unref(m->jobs_in_progress_event_source);
1106         sd_event_source_unref(m->run_queue_event_source);
1107         sd_event_source_unref(m->user_lookup_event_source);
1108
1109         safe_close(m->signal_fd);
1110         safe_close(m->notify_fd);
1111         safe_close(m->cgroups_agent_fd);
1112         safe_close(m->time_change_fd);
1113         safe_close_pair(m->user_lookup_fds);
1114
1115         manager_close_ask_password(m);
1116
1117         manager_close_idle_pipe(m);
1118
1119         udev_unref(m->udev);
1120         sd_event_unref(m->event);
1121
1122         free(m->notify_socket);
1123
1124         lookup_paths_free(&m->lookup_paths);
1125         strv_free(m->environment);
1126
1127         hashmap_free(m->cgroup_unit);
1128         set_free_free(m->unit_path_cache);
1129
1130         free(m->switch_root);
1131         free(m->switch_root_init);
1132
1133         for (i = 0; i < _RLIMIT_MAX; i++)
1134                 m->rlimit[i] = mfree(m->rlimit[i]);
1135
1136         assert(hashmap_isempty(m->units_requiring_mounts_for));
1137         hashmap_free(m->units_requiring_mounts_for);
1138
1139         hashmap_free(m->uid_refs);
1140         hashmap_free(m->gid_refs);
1141
1142         return mfree(m);
1143 }
1144
1145 void manager_enumerate(Manager *m) {
1146         UnitType c;
1147
1148         assert(m);
1149
1150         /* Let's ask every type to load all units from disk/kernel
1151          * that it might know */
1152         for (c = 0; c < _UNIT_TYPE_MAX; c++) {
1153                 if (!unit_type_supported(c)) {
1154                         log_debug("Unit type .%s is not supported on this system.", unit_type_to_string(c));
1155                         continue;
1156                 }
1157
1158                 if (!unit_vtable[c]->enumerate)
1159                         continue;
1160
1161                 unit_vtable[c]->enumerate(m);
1162         }
1163
1164         manager_dispatch_load_queue(m);
1165 }
1166
1167 static void manager_coldplug(Manager *m) {
1168         Iterator i;
1169         Unit *u;
1170         char *k;
1171         int r;
1172
1173         assert(m);
1174
1175         /* Then, let's set up their initial state. */
1176         HASHMAP_FOREACH_KEY(u, k, m->units, i) {
1177
1178                 /* ignore aliases */
1179                 if (u->id != k)
1180                         continue;
1181
1182                 r = unit_coldplug(u);
1183                 if (r < 0)
1184                         log_warning_errno(r, "We couldn't coldplug %s, proceeding anyway: %m", u->id);
1185         }
1186 }
1187
1188 static void manager_build_unit_path_cache(Manager *m) {
1189         char **i;
1190         int r;
1191
1192         assert(m);
1193
1194         set_free_free(m->unit_path_cache);
1195
1196         m->unit_path_cache = set_new(&string_hash_ops);
1197         if (!m->unit_path_cache) {
1198                 r = -ENOMEM;
1199                 goto fail;
1200         }
1201
1202         /* This simply builds a list of files we know exist, so that
1203          * we don't always have to go to disk */
1204
1205         STRV_FOREACH(i, m->lookup_paths.search_path) {
1206                 _cleanup_closedir_ DIR *d = NULL;
1207                 struct dirent *de;
1208
1209                 d = opendir(*i);
1210                 if (!d) {
1211                         if (errno != ENOENT)
1212                                 log_warning_errno(errno, "Failed to open directory %s, ignoring: %m", *i);
1213                         continue;
1214                 }
1215
1216                 FOREACH_DIRENT(de, d, r = -errno; goto fail) {
1217                         char *p;
1218
1219                         p = strjoin(streq(*i, "/") ? "" : *i, "/", de->d_name);
1220                         if (!p) {
1221                                 r = -ENOMEM;
1222                                 goto fail;
1223                         }
1224
1225                         r = set_consume(m->unit_path_cache, p);
1226                         if (r < 0)
1227                                 goto fail;
1228                 }
1229         }
1230
1231         return;
1232
1233 fail:
1234         log_warning_errno(r, "Failed to build unit path cache, proceeding without: %m");
1235         m->unit_path_cache = set_free_free(m->unit_path_cache);
1236 }
1237
1238 static void manager_distribute_fds(Manager *m, FDSet *fds) {
1239         Iterator i;
1240         Unit *u;
1241
1242         assert(m);
1243
1244         HASHMAP_FOREACH(u, m->units, i) {
1245
1246                 if (fdset_size(fds) <= 0)
1247                         break;
1248
1249                 if (!UNIT_VTABLE(u)->distribute_fds)
1250                         continue;
1251
1252                 UNIT_VTABLE(u)->distribute_fds(u, fds);
1253         }
1254 }
1255
1256 int manager_startup(Manager *m, FILE *serialization, FDSet *fds) {
1257         int r, q;
1258
1259         assert(m);
1260
1261         r = lookup_paths_init(&m->lookup_paths, m->unit_file_scope, 0, NULL);
1262         if (r < 0)
1263                 return r;
1264
1265         /* Make sure the transient directory always exists, so that it remains in the search path */
1266         if (!m->test_run) {
1267                 r = mkdir_p_label(m->lookup_paths.transient, 0755);
1268                 if (r < 0)
1269                         return r;
1270         }
1271
1272         dual_timestamp_get(&m->generators_start_timestamp);
1273         r = manager_run_generators(m);
1274         dual_timestamp_get(&m->generators_finish_timestamp);
1275         if (r < 0)
1276                 return r;
1277
1278         lookup_paths_reduce(&m->lookup_paths);
1279         manager_build_unit_path_cache(m);
1280
1281         /* If we will deserialize make sure that during enumeration
1282          * this is already known, so we increase the counter here
1283          * already */
1284         if (serialization)
1285                 m->n_reloading++;
1286
1287         /* First, enumerate what we can from all config files */
1288         dual_timestamp_get(&m->units_load_start_timestamp);
1289         manager_enumerate(m);
1290         dual_timestamp_get(&m->units_load_finish_timestamp);
1291
1292         /* Second, deserialize if there is something to deserialize */
1293         if (serialization)
1294                 r = manager_deserialize(m, serialization, fds);
1295
1296         /* Any fds left? Find some unit which wants them. This is
1297          * useful to allow container managers to pass some file
1298          * descriptors to us pre-initialized. This enables
1299          * socket-based activation of entire containers. */
1300         manager_distribute_fds(m, fds);
1301
1302         /* We might have deserialized the notify fd, but if we didn't
1303          * then let's create the bus now */
1304         q = manager_setup_notify(m);
1305         if (q < 0 && r == 0)
1306                 r = q;
1307
1308         q = manager_setup_cgroups_agent(m);
1309         if (q < 0 && r == 0)
1310                 r = q;
1311
1312         q = manager_setup_user_lookup_fd(m);
1313         if (q < 0 && r == 0)
1314                 r = q;
1315
1316         /* Let's connect to the bus now. */
1317         (void) manager_connect_bus(m, !!serialization);
1318
1319         (void) bus_track_coldplug(m, &m->subscribed, false, m->deserialized_subscribed);
1320         m->deserialized_subscribed = strv_free(m->deserialized_subscribed);
1321
1322         /* Third, fire things up! */
1323         manager_coldplug(m);
1324
1325         /* Release any dynamic users no longer referenced */
1326         dynamic_user_vacuum(m, true);
1327
1328         /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
1329         manager_vacuum_uid_refs(m);
1330         manager_vacuum_gid_refs(m);
1331
1332         if (serialization) {
1333                 assert(m->n_reloading > 0);
1334                 m->n_reloading--;
1335
1336                 /* Let's wait for the UnitNew/JobNew messages being
1337                  * sent, before we notify that the reload is
1338                  * finished */
1339                 m->send_reloading_done = true;
1340         }
1341
1342         return r;
1343 }
1344
1345 int manager_add_job(Manager *m, JobType type, Unit *unit, JobMode mode, sd_bus_error *e, Job **_ret) {
1346         int r;
1347         Transaction *tr;
1348
1349         assert(m);
1350         assert(type < _JOB_TYPE_MAX);
1351         assert(unit);
1352         assert(mode < _JOB_MODE_MAX);
1353
1354         if (mode == JOB_ISOLATE && type != JOB_START)
1355                 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Isolate is only valid for start.");
1356
1357         if (mode == JOB_ISOLATE && !unit->allow_isolate)
1358                 return sd_bus_error_setf(e, BUS_ERROR_NO_ISOLATION, "Operation refused, unit may not be isolated.");
1359
1360         log_unit_debug(unit, "Trying to enqueue job %s/%s/%s", unit->id, job_type_to_string(type), job_mode_to_string(mode));
1361
1362         type = job_type_collapse(type, unit);
1363
1364         tr = transaction_new(mode == JOB_REPLACE_IRREVERSIBLY);
1365         if (!tr)
1366                 return -ENOMEM;
1367
1368         r = transaction_add_job_and_dependencies(tr, type, unit, NULL, true, false,
1369                                                  mode == JOB_IGNORE_DEPENDENCIES || mode == JOB_IGNORE_REQUIREMENTS,
1370                                                  mode == JOB_IGNORE_DEPENDENCIES, e);
1371         if (r < 0)
1372                 goto tr_abort;
1373
1374         if (mode == JOB_ISOLATE) {
1375                 r = transaction_add_isolate_jobs(tr, m);
1376                 if (r < 0)
1377                         goto tr_abort;
1378         }
1379
1380         r = transaction_activate(tr, m, mode, e);
1381         if (r < 0)
1382                 goto tr_abort;
1383
1384         log_unit_debug(unit,
1385                        "Enqueued job %s/%s as %u", unit->id,
1386                        job_type_to_string(type), (unsigned) tr->anchor_job->id);
1387
1388         if (_ret)
1389                 *_ret = tr->anchor_job;
1390
1391         transaction_free(tr);
1392         return 0;
1393
1394 tr_abort:
1395         transaction_abort(tr);
1396         transaction_free(tr);
1397         return r;
1398 }
1399
1400 int manager_add_job_by_name(Manager *m, JobType type, const char *name, JobMode mode, sd_bus_error *e, Job **ret) {
1401         Unit *unit;
1402         int r;
1403
1404         assert(m);
1405         assert(type < _JOB_TYPE_MAX);
1406         assert(name);
1407         assert(mode < _JOB_MODE_MAX);
1408
1409         r = manager_load_unit(m, name, NULL, NULL, &unit);
1410         if (r < 0)
1411                 return r;
1412
1413         return manager_add_job(m, type, unit, mode, e, ret);
1414 }
1415
1416 int manager_add_job_by_name_and_warn(Manager *m, JobType type, const char *name, JobMode mode, Job **ret) {
1417         _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1418         int r;
1419
1420         assert(m);
1421         assert(type < _JOB_TYPE_MAX);
1422         assert(name);
1423         assert(mode < _JOB_MODE_MAX);
1424
1425         r = manager_add_job_by_name(m, type, name, mode, &error, ret);
1426         if (r < 0)
1427                 return log_warning_errno(r, "Failed to enqueue %s job for %s: %s", job_mode_to_string(mode), name, bus_error_message(&error, r));
1428
1429         return r;
1430 }
1431
1432 Job *manager_get_job(Manager *m, uint32_t id) {
1433         assert(m);
1434
1435         return hashmap_get(m->jobs, UINT32_TO_PTR(id));
1436 }
1437
1438 Unit *manager_get_unit(Manager *m, const char *name) {
1439         assert(m);
1440         assert(name);
1441
1442         return hashmap_get(m->units, name);
1443 }
1444
1445 unsigned manager_dispatch_load_queue(Manager *m) {
1446         Unit *u;
1447         unsigned n = 0;
1448
1449         assert(m);
1450
1451         /* Make sure we are not run recursively */
1452         if (m->dispatching_load_queue)
1453                 return 0;
1454
1455         m->dispatching_load_queue = true;
1456
1457         /* Dispatches the load queue. Takes a unit from the queue and
1458          * tries to load its data until the queue is empty */
1459
1460         while ((u = m->load_queue)) {
1461                 assert(u->in_load_queue);
1462
1463                 unit_load(u);
1464                 n++;
1465         }
1466
1467         m->dispatching_load_queue = false;
1468         return n;
1469 }
1470
1471 int manager_load_unit_prepare(
1472                 Manager *m,
1473                 const char *name,
1474                 const char *path,
1475                 sd_bus_error *e,
1476                 Unit **_ret) {
1477
1478         Unit *ret;
1479         UnitType t;
1480         int r;
1481
1482         assert(m);
1483         assert(name || path);
1484
1485         /* This will prepare the unit for loading, but not actually
1486          * load anything from disk. */
1487
1488         if (path && !is_path(path))
1489                 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Path %s is not absolute.", path);
1490
1491         if (!name)
1492                 name = basename(path);
1493
1494         t = unit_name_to_type(name);
1495
1496         if (t == _UNIT_TYPE_INVALID || !unit_name_is_valid(name, UNIT_NAME_PLAIN|UNIT_NAME_INSTANCE)) {
1497                 if (unit_name_is_valid(name, UNIT_NAME_TEMPLATE))
1498                         return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is missing the instance name.", name);
1499
1500                 return sd_bus_error_setf(e, SD_BUS_ERROR_INVALID_ARGS, "Unit name %s is not valid.", name);
1501         }
1502
1503         ret = manager_get_unit(m, name);
1504         if (ret) {
1505                 *_ret = ret;
1506                 return 1;
1507         }
1508
1509         ret = unit_new(m, unit_vtable[t]->object_size);
1510         if (!ret)
1511                 return -ENOMEM;
1512
1513         if (path) {
1514                 ret->fragment_path = strdup(path);
1515                 if (!ret->fragment_path) {
1516                         unit_free(ret);
1517                         return -ENOMEM;
1518                 }
1519         }
1520
1521         r = unit_add_name(ret, name);
1522         if (r < 0) {
1523                 unit_free(ret);
1524                 return r;
1525         }
1526
1527         unit_add_to_load_queue(ret);
1528         unit_add_to_dbus_queue(ret);
1529         unit_add_to_gc_queue(ret);
1530
1531         if (_ret)
1532                 *_ret = ret;
1533
1534         return 0;
1535 }
1536
1537 int manager_load_unit(
1538                 Manager *m,
1539                 const char *name,
1540                 const char *path,
1541                 sd_bus_error *e,
1542                 Unit **_ret) {
1543
1544         int r;
1545
1546         assert(m);
1547
1548         /* This will load the service information files, but not actually
1549          * start any services or anything. */
1550
1551         r = manager_load_unit_prepare(m, name, path, e, _ret);
1552         if (r != 0)
1553                 return r;
1554
1555         manager_dispatch_load_queue(m);
1556
1557         if (_ret)
1558                 *_ret = unit_follow_merge(*_ret);
1559
1560         return 0;
1561 }
1562
1563 void manager_dump_jobs(Manager *s, FILE *f, const char *prefix) {
1564         Iterator i;
1565         Job *j;
1566
1567         assert(s);
1568         assert(f);
1569
1570         HASHMAP_FOREACH(j, s->jobs, i)
1571                 job_dump(j, f, prefix);
1572 }
1573
1574 void manager_dump_units(Manager *s, FILE *f, const char *prefix) {
1575         Iterator i;
1576         Unit *u;
1577         const char *t;
1578
1579         assert(s);
1580         assert(f);
1581
1582         HASHMAP_FOREACH_KEY(u, t, s->units, i)
1583                 if (u->id == t)
1584                         unit_dump(u, f, prefix);
1585 }
1586
1587 void manager_clear_jobs(Manager *m) {
1588         Job *j;
1589
1590         assert(m);
1591
1592         while ((j = hashmap_first(m->jobs)))
1593                 /* No need to recurse. We're cancelling all jobs. */
1594                 job_finish_and_invalidate(j, JOB_CANCELED, false, false);
1595 }
1596
1597 static int manager_dispatch_run_queue(sd_event_source *source, void *userdata) {
1598         Manager *m = userdata;
1599         Job *j;
1600
1601         assert(source);
1602         assert(m);
1603
1604         while ((j = m->run_queue)) {
1605                 assert(j->installed);
1606                 assert(j->in_run_queue);
1607
1608                 job_run_and_invalidate(j);
1609         }
1610
1611         if (m->n_running_jobs > 0)
1612                 manager_watch_jobs_in_progress(m);
1613
1614         if (m->n_on_console > 0)
1615                 manager_watch_idle_pipe(m);
1616
1617         return 1;
1618 }
1619
1620 static unsigned manager_dispatch_dbus_queue(Manager *m) {
1621         Job *j;
1622         Unit *u;
1623         unsigned n = 0;
1624
1625         assert(m);
1626
1627         if (m->dispatching_dbus_queue)
1628                 return 0;
1629
1630         m->dispatching_dbus_queue = true;
1631
1632         while ((u = m->dbus_unit_queue)) {
1633                 assert(u->in_dbus_queue);
1634
1635                 bus_unit_send_change_signal(u);
1636                 n++;
1637         }
1638
1639         while ((j = m->dbus_job_queue)) {
1640                 assert(j->in_dbus_queue);
1641
1642                 bus_job_send_change_signal(j);
1643                 n++;
1644         }
1645
1646         m->dispatching_dbus_queue = false;
1647
1648         if (m->send_reloading_done) {
1649                 m->send_reloading_done = false;
1650
1651                 bus_manager_send_reloading(m, false);
1652         }
1653
1654         if (m->queued_message)
1655                 bus_send_queued_message(m);
1656
1657         return n;
1658 }
1659
1660 static int manager_dispatch_cgroups_agent_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1661         Manager *m = userdata;
1662         char buf[PATH_MAX+1];
1663         ssize_t n;
1664
1665         n = recv(fd, buf, sizeof(buf), 0);
1666         if (n < 0)
1667                 return log_error_errno(errno, "Failed to read cgroups agent message: %m");
1668         if (n == 0) {
1669                 log_error("Got zero-length cgroups agent message, ignoring.");
1670                 return 0;
1671         }
1672         if ((size_t) n >= sizeof(buf)) {
1673                 log_error("Got overly long cgroups agent message, ignoring.");
1674                 return 0;
1675         }
1676
1677         if (memchr(buf, 0, n)) {
1678                 log_error("Got cgroups agent message with embedded NUL byte, ignoring.");
1679                 return 0;
1680         }
1681         buf[n] = 0;
1682
1683         manager_notify_cgroup_empty(m, buf);
1684         bus_forward_agent_released(m, buf);
1685
1686         return 0;
1687 }
1688
1689 static void manager_invoke_notify_message(Manager *m, Unit *u, pid_t pid, const char *buf, FDSet *fds) {
1690         _cleanup_strv_free_ char **tags = NULL;
1691
1692         assert(m);
1693         assert(u);
1694         assert(buf);
1695
1696         tags = strv_split(buf, "\n\r");
1697         if (!tags) {
1698                 log_oom();
1699                 return;
1700         }
1701
1702         if (UNIT_VTABLE(u)->notify_message)
1703                 UNIT_VTABLE(u)->notify_message(u, pid, tags, fds);
1704         else if (_unlikely_(log_get_max_level() >= LOG_DEBUG)) {
1705                 _cleanup_free_ char *x = NULL, *y = NULL;
1706
1707                 x = cescape(buf);
1708                 if (x)
1709                         y = ellipsize(x, 20, 90);
1710                 log_unit_debug(u, "Got notification message \"%s\", ignoring.", strnull(y));
1711         }
1712 }
1713
1714 static int manager_dispatch_notify_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1715
1716         _cleanup_fdset_free_ FDSet *fds = NULL;
1717         Manager *m = userdata;
1718         char buf[NOTIFY_BUFFER_MAX+1];
1719         struct iovec iovec = {
1720                 .iov_base = buf,
1721                 .iov_len = sizeof(buf)-1,
1722         };
1723         union {
1724                 struct cmsghdr cmsghdr;
1725                 uint8_t buf[CMSG_SPACE(sizeof(struct ucred)) +
1726                             CMSG_SPACE(sizeof(int) * NOTIFY_FD_MAX)];
1727         } control = {};
1728         struct msghdr msghdr = {
1729                 .msg_iov = &iovec,
1730                 .msg_iovlen = 1,
1731                 .msg_control = &control,
1732                 .msg_controllen = sizeof(control),
1733         };
1734
1735         struct cmsghdr *cmsg;
1736         struct ucred *ucred = NULL;
1737         Unit *u1, *u2, *u3;
1738         int r, *fd_array = NULL;
1739         unsigned n_fds = 0;
1740         ssize_t n;
1741
1742         assert(m);
1743         assert(m->notify_fd == fd);
1744
1745         if (revents != EPOLLIN) {
1746                 log_warning("Got unexpected poll event for notify fd.");
1747                 return 0;
1748         }
1749
1750         n = recvmsg(m->notify_fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC|MSG_TRUNC);
1751         if (n < 0) {
1752                 if (IN_SET(errno, EAGAIN, EINTR))
1753                         return 0; /* Spurious wakeup, try again */
1754
1755                 /* If this is any other, real error, then let's stop processing this socket. This of course means we
1756                  * won't take notification messages anymore, but that's still better than busy looping around this:
1757                  * being woken up over and over again but being unable to actually read the message off the socket. */
1758                 return log_error_errno(errno, "Failed to receive notification message: %m");
1759         }
1760
1761         CMSG_FOREACH(cmsg, &msghdr) {
1762                 if (cmsg->cmsg_level == SOL_SOCKET && cmsg->cmsg_type == SCM_RIGHTS) {
1763
1764                         fd_array = (int*) CMSG_DATA(cmsg);
1765                         n_fds = (cmsg->cmsg_len - CMSG_LEN(0)) / sizeof(int);
1766
1767                 } else if (cmsg->cmsg_level == SOL_SOCKET &&
1768                            cmsg->cmsg_type == SCM_CREDENTIALS &&
1769                            cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
1770
1771                         ucred = (struct ucred*) CMSG_DATA(cmsg);
1772                 }
1773         }
1774
1775         if (n_fds > 0) {
1776                 assert(fd_array);
1777
1778                 r = fdset_new_array(&fds, fd_array, n_fds);
1779                 if (r < 0) {
1780                         close_many(fd_array, n_fds);
1781                         log_oom();
1782                         return 0;
1783                 }
1784         }
1785
1786         if (!ucred || ucred->pid <= 0) {
1787                 log_warning("Received notify message without valid credentials. Ignoring.");
1788                 return 0;
1789         }
1790
1791         if ((size_t) n >= sizeof(buf) || (msghdr.msg_flags & MSG_TRUNC)) {
1792                 log_warning("Received notify message exceeded maximum size. Ignoring.");
1793                 return 0;
1794         }
1795
1796         /* As extra safety check, let's make sure the string we get doesn't contain embedded NUL bytes. We permit one
1797          * trailing NUL byte in the message, but don't expect it. */
1798         if (n > 1 && memchr(buf, 0, n-1)) {
1799                 log_warning("Received notify message with embedded NUL bytes. Ignoring.");
1800                 return 0;
1801         }
1802
1803         /* Make sure it's NUL-terminated. */
1804         buf[n] = 0;
1805
1806         /* Notify every unit that might be interested, but try
1807          * to avoid notifying the same one multiple times. */
1808         u1 = manager_get_unit_by_pid_cgroup(m, ucred->pid);
1809         if (u1)
1810                 manager_invoke_notify_message(m, u1, ucred->pid, buf, fds);
1811
1812         u2 = hashmap_get(m->watch_pids1, PID_TO_PTR(ucred->pid));
1813         if (u2 && u2 != u1)
1814                 manager_invoke_notify_message(m, u2, ucred->pid, buf, fds);
1815
1816         u3 = hashmap_get(m->watch_pids2, PID_TO_PTR(ucred->pid));
1817         if (u3 && u3 != u2 && u3 != u1)
1818                 manager_invoke_notify_message(m, u3, ucred->pid, buf, fds);
1819
1820         if (!u1 && !u2 && !u3)
1821                 log_warning("Cannot find unit for notify message of PID "PID_FMT".", ucred->pid);
1822
1823         if (fdset_size(fds) > 0)
1824                 log_warning("Got extra auxiliary fds with notification message, closing them.");
1825
1826         return 0;
1827 }
1828
1829 static void invoke_sigchld_event(Manager *m, Unit *u, const siginfo_t *si) {
1830         uint64_t iteration;
1831
1832         assert(m);
1833         assert(u);
1834         assert(si);
1835
1836         sd_event_get_iteration(m->event, &iteration);
1837
1838         log_unit_debug(u, "Child "PID_FMT" belongs to %s", si->si_pid, u->id);
1839
1840         unit_unwatch_pid(u, si->si_pid);
1841
1842         if (UNIT_VTABLE(u)->sigchld_event) {
1843                 if (set_size(u->pids) <= 1 ||
1844                     iteration != u->sigchldgen ||
1845                     unit_main_pid(u) == si->si_pid ||
1846                     unit_control_pid(u) == si->si_pid) {
1847                         UNIT_VTABLE(u)->sigchld_event(u, si->si_pid, si->si_code, si->si_status);
1848                         u->sigchldgen = iteration;
1849                 } else
1850                         log_debug("%s already issued a sigchld this iteration %" PRIu64 ", skipping. Pids still being watched %d", u->id, iteration, set_size(u->pids));
1851          }
1852 }
1853
1854 static int manager_dispatch_sigchld(Manager *m) {
1855         assert(m);
1856
1857         for (;;) {
1858                 siginfo_t si = {};
1859
1860                 /* First we call waitd() for a PID and do not reap the
1861                  * zombie. That way we can still access /proc/$PID for
1862                  * it while it is a zombie. */
1863                 if (waitid(P_ALL, 0, &si, WEXITED|WNOHANG|WNOWAIT) < 0) {
1864
1865                         if (errno == ECHILD)
1866                                 break;
1867
1868                         if (errno == EINTR)
1869                                 continue;
1870
1871                         return -errno;
1872                 }
1873
1874                 if (si.si_pid <= 0)
1875                         break;
1876
1877                 if (si.si_code == CLD_EXITED || si.si_code == CLD_KILLED || si.si_code == CLD_DUMPED) {
1878                         _cleanup_free_ char *name = NULL;
1879                         Unit *u1, *u2, *u3;
1880
1881                         get_process_comm(si.si_pid, &name);
1882
1883                         log_debug("Child "PID_FMT" (%s) died (code=%s, status=%i/%s)",
1884                                   si.si_pid, strna(name),
1885                                   sigchld_code_to_string(si.si_code),
1886                                   si.si_status,
1887                                   strna(si.si_code == CLD_EXITED
1888                                         ? exit_status_to_string(si.si_status, EXIT_STATUS_FULL)
1889                                         : signal_to_string(si.si_status)));
1890
1891                         /* And now figure out the unit this belongs
1892                          * to, it might be multiple... */
1893                         u1 = manager_get_unit_by_pid_cgroup(m, si.si_pid);
1894                         if (u1)
1895                                 invoke_sigchld_event(m, u1, &si);
1896                         u2 = hashmap_get(m->watch_pids1, PID_TO_PTR(si.si_pid));
1897                         if (u2 && u2 != u1)
1898                                 invoke_sigchld_event(m, u2, &si);
1899                         u3 = hashmap_get(m->watch_pids2, PID_TO_PTR(si.si_pid));
1900                         if (u3 && u3 != u2 && u3 != u1)
1901                                 invoke_sigchld_event(m, u3, &si);
1902                 }
1903
1904                 /* And now, we actually reap the zombie. */
1905                 if (waitid(P_PID, si.si_pid, &si, WEXITED) < 0) {
1906                         if (errno == EINTR)
1907                                 continue;
1908
1909                         return -errno;
1910                 }
1911         }
1912
1913         return 0;
1914 }
1915
1916 static int manager_start_target(Manager *m, const char *name, JobMode mode) {
1917         _cleanup_(sd_bus_error_free) sd_bus_error error = SD_BUS_ERROR_NULL;
1918         int r;
1919
1920         log_debug("Activating special unit %s", name);
1921
1922         r = manager_add_job_by_name(m, JOB_START, name, mode, &error, NULL);
1923         if (r < 0)
1924                 log_error("Failed to enqueue %s job: %s", name, bus_error_message(&error, r));
1925
1926         return r;
1927 }
1928
1929 static void manager_handle_ctrl_alt_del(Manager *m) {
1930         /* If the user presses C-A-D more than
1931          * 7 times within 2s, we reboot/shutdown immediately,
1932          * unless it was disabled in system.conf */
1933
1934         if (ratelimit_test(&m->ctrl_alt_del_ratelimit) || m->cad_burst_action == EMERGENCY_ACTION_NONE)
1935                 manager_start_target(m, SPECIAL_CTRL_ALT_DEL_TARGET, JOB_REPLACE_IRREVERSIBLY);
1936         else
1937                 emergency_action(m, m->cad_burst_action, NULL,
1938                                 "Ctrl-Alt-Del was pressed more than 7 times within 2s");
1939 }
1940
1941 static int manager_dispatch_signal_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
1942         Manager *m = userdata;
1943         ssize_t n;
1944         struct signalfd_siginfo sfsi;
1945         bool sigchld = false;
1946         int r;
1947
1948         assert(m);
1949         assert(m->signal_fd == fd);
1950
1951         if (revents != EPOLLIN) {
1952                 log_warning("Got unexpected events from signal file descriptor.");
1953                 return 0;
1954         }
1955
1956         for (;;) {
1957                 n = read(m->signal_fd, &sfsi, sizeof(sfsi));
1958                 if (n != sizeof(sfsi)) {
1959                         if (n >= 0) {
1960                                 log_warning("Truncated read from signal fd (%zu bytes)!", n);
1961                                 return 0;
1962                         }
1963
1964                         if (IN_SET(errno, EINTR, EAGAIN))
1965                                 break;
1966
1967                         /* We return an error here, which will kill this handler,
1968                          * to avoid a busy loop on read error. */
1969                         return log_error_errno(errno, "Reading from signal fd failed: %m");
1970                 }
1971
1972                 log_received_signal(sfsi.ssi_signo == SIGCHLD ||
1973                                     (sfsi.ssi_signo == SIGTERM && MANAGER_IS_USER(m))
1974                                     ? LOG_DEBUG : LOG_INFO,
1975                                     &sfsi);
1976
1977                 switch (sfsi.ssi_signo) {
1978
1979                 case SIGCHLD:
1980                         sigchld = true;
1981                         break;
1982
1983                 case SIGTERM:
1984                         if (MANAGER_IS_SYSTEM(m)) {
1985                                 /* This is for compatibility with the
1986                                  * original sysvinit */
1987                                 m->exit_code = MANAGER_REEXECUTE;
1988                                 break;
1989                         }
1990
1991                         /* Fall through */
1992
1993                 case SIGINT:
1994                         if (MANAGER_IS_SYSTEM(m)) {
1995                                 manager_handle_ctrl_alt_del(m);
1996                                 break;
1997                         }
1998
1999                         /* Run the exit target if there is one, if not, just exit. */
2000                         if (manager_start_target(m, SPECIAL_EXIT_TARGET, JOB_REPLACE) < 0) {
2001                                 m->exit_code = MANAGER_EXIT;
2002                                 return 0;
2003                         }
2004
2005                         break;
2006
2007                 case SIGWINCH:
2008                         if (MANAGER_IS_SYSTEM(m))
2009                                 manager_start_target(m, SPECIAL_KBREQUEST_TARGET, JOB_REPLACE);
2010
2011                         /* This is a nop on non-init */
2012                         break;
2013
2014                 case SIGPWR:
2015                         if (MANAGER_IS_SYSTEM(m))
2016                                 manager_start_target(m, SPECIAL_SIGPWR_TARGET, JOB_REPLACE);
2017
2018                         /* This is a nop on non-init */
2019                         break;
2020
2021                 case SIGUSR1: {
2022                         Unit *u;
2023
2024                         u = manager_get_unit(m, SPECIAL_DBUS_SERVICE);
2025
2026                         if (!u || UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u))) {
2027                                 log_info("Trying to reconnect to bus...");
2028                                 bus_init(m, true);
2029                         }
2030
2031                         if (!u || !UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u))) {
2032                                 log_info("Loading D-Bus service...");
2033                                 manager_start_target(m, SPECIAL_DBUS_SERVICE, JOB_REPLACE);
2034                         }
2035
2036                         break;
2037                 }
2038
2039                 case SIGUSR2: {
2040                         _cleanup_free_ char *dump = NULL;
2041                         _cleanup_fclose_ FILE *f = NULL;
2042                         size_t size;
2043
2044                         f = open_memstream(&dump, &size);
2045                         if (!f) {
2046                                 log_warning_errno(errno, "Failed to allocate memory stream: %m");
2047                                 break;
2048                         }
2049
2050                         manager_dump_units(m, f, "\t");
2051                         manager_dump_jobs(m, f, "\t");
2052
2053                         r = fflush_and_check(f);
2054                         if (r < 0) {
2055                                 log_warning_errno(r, "Failed to write status stream: %m");
2056                                 break;
2057                         }
2058
2059                         log_dump(LOG_INFO, dump);
2060                         break;
2061                 }
2062
2063                 case SIGHUP:
2064                         m->exit_code = MANAGER_RELOAD;
2065                         break;
2066
2067                 default: {
2068
2069                         /* Starting SIGRTMIN+0 */
2070                         static const char * const target_table[] = {
2071                                 [0] = SPECIAL_DEFAULT_TARGET,
2072                                 [1] = SPECIAL_RESCUE_TARGET,
2073                                 [2] = SPECIAL_EMERGENCY_TARGET,
2074                                 [3] = SPECIAL_HALT_TARGET,
2075                                 [4] = SPECIAL_POWEROFF_TARGET,
2076                                 [5] = SPECIAL_REBOOT_TARGET,
2077                                 [6] = SPECIAL_KEXEC_TARGET
2078                         };
2079
2080                         /* Starting SIGRTMIN+13, so that target halt and system halt are 10 apart */
2081                         static const ManagerExitCode code_table[] = {
2082                                 [0] = MANAGER_HALT,
2083                                 [1] = MANAGER_POWEROFF,
2084                                 [2] = MANAGER_REBOOT,
2085                                 [3] = MANAGER_KEXEC
2086                         };
2087
2088                         if ((int) sfsi.ssi_signo >= SIGRTMIN+0 &&
2089                             (int) sfsi.ssi_signo < SIGRTMIN+(int) ELEMENTSOF(target_table)) {
2090                                 int idx = (int) sfsi.ssi_signo - SIGRTMIN;
2091                                 manager_start_target(m, target_table[idx],
2092                                                      (idx == 1 || idx == 2) ? JOB_ISOLATE : JOB_REPLACE);
2093                                 break;
2094                         }
2095
2096                         if ((int) sfsi.ssi_signo >= SIGRTMIN+13 &&
2097                             (int) sfsi.ssi_signo < SIGRTMIN+13+(int) ELEMENTSOF(code_table)) {
2098                                 m->exit_code = code_table[sfsi.ssi_signo - SIGRTMIN - 13];
2099                                 break;
2100                         }
2101
2102                         switch (sfsi.ssi_signo - SIGRTMIN) {
2103
2104                         case 20:
2105                                 manager_set_show_status(m, SHOW_STATUS_YES);
2106                                 break;
2107
2108                         case 21:
2109                                 manager_set_show_status(m, SHOW_STATUS_NO);
2110                                 break;
2111
2112                         case 22:
2113                                 log_set_max_level(LOG_DEBUG);
2114                                 log_info("Setting log level to debug.");
2115                                 break;
2116
2117                         case 23:
2118                                 log_set_max_level(LOG_INFO);
2119                                 log_info("Setting log level to info.");
2120                                 break;
2121
2122                         case 24:
2123                                 if (MANAGER_IS_USER(m)) {
2124                                         m->exit_code = MANAGER_EXIT;
2125                                         return 0;
2126                                 }
2127
2128                                 /* This is a nop on init */
2129                                 break;
2130
2131                         case 26:
2132                         case 29: /* compatibility: used to be mapped to LOG_TARGET_SYSLOG_OR_KMSG */
2133                                 log_set_target(LOG_TARGET_JOURNAL_OR_KMSG);
2134                                 log_notice("Setting log target to journal-or-kmsg.");
2135                                 break;
2136
2137                         case 27:
2138                                 log_set_target(LOG_TARGET_CONSOLE);
2139                                 log_notice("Setting log target to console.");
2140                                 break;
2141
2142                         case 28:
2143                                 log_set_target(LOG_TARGET_KMSG);
2144                                 log_notice("Setting log target to kmsg.");
2145                                 break;
2146
2147                         default:
2148                                 log_warning("Got unhandled signal <%s>.", signal_to_string(sfsi.ssi_signo));
2149                         }
2150                 }
2151                 }
2152         }
2153
2154         if (sigchld)
2155                 manager_dispatch_sigchld(m);
2156
2157         return 0;
2158 }
2159
2160 static int manager_dispatch_time_change_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2161         Manager *m = userdata;
2162         Iterator i;
2163         Unit *u;
2164
2165         assert(m);
2166         assert(m->time_change_fd == fd);
2167
2168         log_struct(LOG_INFO,
2169                    LOG_MESSAGE_ID(SD_MESSAGE_TIME_CHANGE),
2170                    LOG_MESSAGE("Time has been changed"),
2171                    NULL);
2172
2173         /* Restart the watch */
2174         m->time_change_event_source = sd_event_source_unref(m->time_change_event_source);
2175         m->time_change_fd = safe_close(m->time_change_fd);
2176
2177         manager_setup_time_change(m);
2178
2179         HASHMAP_FOREACH(u, m->units, i)
2180                 if (UNIT_VTABLE(u)->time_change)
2181                         UNIT_VTABLE(u)->time_change(u);
2182
2183         return 0;
2184 }
2185
2186 static int manager_dispatch_idle_pipe_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
2187         Manager *m = userdata;
2188
2189         assert(m);
2190         assert(m->idle_pipe[2] == fd);
2191
2192         m->no_console_output = m->n_on_console > 0;
2193
2194         manager_close_idle_pipe(m);
2195
2196         return 0;
2197 }
2198
2199 static int manager_dispatch_jobs_in_progress(sd_event_source *source, usec_t usec, void *userdata) {
2200         Manager *m = userdata;
2201         int r;
2202         uint64_t next;
2203
2204         assert(m);
2205         assert(source);
2206
2207         manager_print_jobs_in_progress(m);
2208
2209         next = now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_PERIOD_USEC;
2210         r = sd_event_source_set_time(source, next);
2211         if (r < 0)
2212                 return r;
2213
2214         return sd_event_source_set_enabled(source, SD_EVENT_ONESHOT);
2215 }
2216
2217 int manager_loop(Manager *m) {
2218         int r;
2219
2220         RATELIMIT_DEFINE(rl, 1*USEC_PER_SEC, 50000);
2221
2222         assert(m);
2223         m->exit_code = MANAGER_OK;
2224
2225         /* Release the path cache */
2226         m->unit_path_cache = set_free_free(m->unit_path_cache);
2227
2228         manager_check_finished(m);
2229
2230         /* There might still be some zombies hanging around from
2231          * before we were exec()'ed. Let's reap them. */
2232         r = manager_dispatch_sigchld(m);
2233         if (r < 0)
2234                 return r;
2235
2236         while (m->exit_code == MANAGER_OK) {
2237                 usec_t wait_usec;
2238
2239                 if (m->runtime_watchdog > 0 && m->runtime_watchdog != USEC_INFINITY && MANAGER_IS_SYSTEM(m))
2240                         watchdog_ping();
2241
2242                 if (!ratelimit_test(&rl)) {
2243                         /* Yay, something is going seriously wrong, pause a little */
2244                         log_warning("Looping too fast. Throttling execution a little.");
2245                         sleep(1);
2246                 }
2247
2248                 if (manager_dispatch_load_queue(m) > 0)
2249                         continue;
2250
2251                 if (manager_dispatch_gc_job_queue(m) > 0)
2252                         continue;
2253
2254                 if (manager_dispatch_gc_unit_queue(m) > 0)
2255                         continue;
2256
2257                 if (manager_dispatch_cleanup_queue(m) > 0)
2258                         continue;
2259
2260                 if (manager_dispatch_cgroup_queue(m) > 0)
2261                         continue;
2262
2263                 if (manager_dispatch_dbus_queue(m) > 0)
2264                         continue;
2265
2266                 /* Sleep for half the watchdog time */
2267                 if (m->runtime_watchdog > 0 && m->runtime_watchdog != USEC_INFINITY && MANAGER_IS_SYSTEM(m)) {
2268                         wait_usec = m->runtime_watchdog / 2;
2269                         if (wait_usec <= 0)
2270                                 wait_usec = 1;
2271                 } else
2272                         wait_usec = USEC_INFINITY;
2273
2274                 r = sd_event_run(m->event, wait_usec);
2275                 if (r < 0)
2276                         return log_error_errno(r, "Failed to run event loop: %m");
2277         }
2278
2279         return m->exit_code;
2280 }
2281
2282 int manager_load_unit_from_dbus_path(Manager *m, const char *s, sd_bus_error *e, Unit **_u) {
2283         _cleanup_free_ char *n = NULL;
2284         sd_id128_t invocation_id;
2285         Unit *u;
2286         int r;
2287
2288         assert(m);
2289         assert(s);
2290         assert(_u);
2291
2292         r = unit_name_from_dbus_path(s, &n);
2293         if (r < 0)
2294                 return r;
2295
2296         /* Permit addressing units by invocation ID: if the passed bus path is suffixed by a 128bit ID then we use it
2297          * as invocation ID. */
2298         r = sd_id128_from_string(n, &invocation_id);
2299         if (r >= 0) {
2300                 u = hashmap_get(m->units_by_invocation_id, &invocation_id);
2301                 if (u) {
2302                         *_u = u;
2303                         return 0;
2304                 }
2305
2306                 return sd_bus_error_setf(e, BUS_ERROR_NO_UNIT_FOR_INVOCATION_ID, "No unit with the specified invocation ID " SD_ID128_FORMAT_STR " known.", SD_ID128_FORMAT_VAL(invocation_id));
2307         }
2308
2309         /* If this didn't work, we use the suffix as unit name. */
2310         r = manager_load_unit(m, n, NULL, e, &u);
2311         if (r < 0)
2312                 return r;
2313
2314         *_u = u;
2315         return 0;
2316 }
2317
2318 int manager_get_job_from_dbus_path(Manager *m, const char *s, Job **_j) {
2319         const char *p;
2320         unsigned id;
2321         Job *j;
2322         int r;
2323
2324         assert(m);
2325         assert(s);
2326         assert(_j);
2327
2328         p = startswith(s, "/org/freedesktop/systemd1/job/");
2329         if (!p)
2330                 return -EINVAL;
2331
2332         r = safe_atou(p, &id);
2333         if (r < 0)
2334                 return r;
2335
2336         j = manager_get_job(m, id);
2337         if (!j)
2338                 return -ENOENT;
2339
2340         *_j = j;
2341
2342         return 0;
2343 }
2344
2345 void manager_send_unit_audit(Manager *m, Unit *u, int type, bool success) {
2346
2347 #ifdef HAVE_AUDIT
2348         _cleanup_free_ char *p = NULL;
2349         const char *msg;
2350         int audit_fd, r;
2351
2352         if (!MANAGER_IS_SYSTEM(m))
2353                 return;
2354
2355         audit_fd = get_audit_fd();
2356         if (audit_fd < 0)
2357                 return;
2358
2359         /* Don't generate audit events if the service was already
2360          * started and we're just deserializing */
2361         if (MANAGER_IS_RELOADING(m))
2362                 return;
2363
2364         if (u->type != UNIT_SERVICE)
2365                 return;
2366
2367         r = unit_name_to_prefix_and_instance(u->id, &p);
2368         if (r < 0) {
2369                 log_error_errno(r, "Failed to extract prefix and instance of unit name: %m");
2370                 return;
2371         }
2372
2373         msg = strjoina("unit=", p);
2374         if (audit_log_user_comm_message(audit_fd, type, msg, "systemd", NULL, NULL, NULL, success) < 0) {
2375                 if (errno == EPERM)
2376                         /* We aren't allowed to send audit messages?
2377                          * Then let's not retry again. */
2378                         close_audit_fd();
2379                 else
2380                         log_warning_errno(errno, "Failed to send audit message: %m");
2381         }
2382 #endif
2383
2384 }
2385
2386 void manager_send_unit_plymouth(Manager *m, Unit *u) {
2387         static const union sockaddr_union sa = PLYMOUTH_SOCKET;
2388         _cleanup_free_ char *message = NULL;
2389         _cleanup_close_ int fd = -1;
2390         int n = 0;
2391
2392         /* Don't generate plymouth events if the service was already
2393          * started and we're just deserializing */
2394         if (MANAGER_IS_RELOADING(m))
2395                 return;
2396
2397         if (!MANAGER_IS_SYSTEM(m))
2398                 return;
2399
2400         if (detect_container() > 0)
2401                 return;
2402
2403         if (u->type != UNIT_SERVICE &&
2404             u->type != UNIT_MOUNT &&
2405             u->type != UNIT_SWAP)
2406                 return;
2407
2408         /* We set SOCK_NONBLOCK here so that we rather drop the
2409          * message then wait for plymouth */
2410         fd = socket(AF_UNIX, SOCK_STREAM|SOCK_CLOEXEC|SOCK_NONBLOCK, 0);
2411         if (fd < 0) {
2412                 log_error_errno(errno, "socket() failed: %m");
2413                 return;
2414         }
2415
2416         if (connect(fd, &sa.sa, SOCKADDR_UN_LEN(sa.un)) < 0) {
2417
2418                 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2419                         log_error_errno(errno, "connect() failed: %m");
2420                 return;
2421         }
2422
2423         if (asprintf(&message, "U\002%c%s%n", (int) (strlen(u->id) + 1), u->id, &n) < 0) {
2424                 log_oom();
2425                 return;
2426         }
2427
2428         errno = 0;
2429         if (write(fd, message, n + 1) != n + 1)
2430                 if (!IN_SET(errno, EPIPE, EAGAIN, ENOENT, ECONNREFUSED, ECONNRESET, ECONNABORTED))
2431                         log_error_errno(errno, "Failed to write Plymouth message: %m");
2432 }
2433
2434 int manager_open_serialization(Manager *m, FILE **_f) {
2435         const char *path;
2436         int fd = -1;
2437         FILE *f;
2438
2439         assert(_f);
2440
2441         path = MANAGER_IS_SYSTEM(m) ? "/run/systemd" : "/tmp";
2442         fd = open_tmpfile_unlinkable(path, O_RDWR|O_CLOEXEC);
2443         if (fd < 0)
2444                 return -errno;
2445
2446         log_debug("Serializing state to %s", path);
2447
2448         f = fdopen(fd, "w+");
2449         if (!f) {
2450                 safe_close(fd);
2451                 return -errno;
2452         }
2453
2454         *_f = f;
2455
2456         return 0;
2457 }
2458
2459 int manager_serialize(Manager *m, FILE *f, FDSet *fds, bool switching_root) {
2460         Iterator i;
2461         Unit *u;
2462         const char *t;
2463         char **e;
2464         int r;
2465
2466         assert(m);
2467         assert(f);
2468         assert(fds);
2469
2470         m->n_reloading++;
2471
2472         fprintf(f, "current-job-id=%"PRIu32"\n", m->current_job_id);
2473         fprintf(f, "taint-usr=%s\n", yes_no(m->taint_usr));
2474         fprintf(f, "n-installed-jobs=%u\n", m->n_installed_jobs);
2475         fprintf(f, "n-failed-jobs=%u\n", m->n_failed_jobs);
2476
2477         dual_timestamp_serialize(f, "firmware-timestamp", &m->firmware_timestamp);
2478         dual_timestamp_serialize(f, "loader-timestamp", &m->loader_timestamp);
2479         dual_timestamp_serialize(f, "kernel-timestamp", &m->kernel_timestamp);
2480         dual_timestamp_serialize(f, "initrd-timestamp", &m->initrd_timestamp);
2481
2482         if (!in_initrd()) {
2483                 dual_timestamp_serialize(f, "userspace-timestamp", &m->userspace_timestamp);
2484                 dual_timestamp_serialize(f, "finish-timestamp", &m->finish_timestamp);
2485                 dual_timestamp_serialize(f, "security-start-timestamp", &m->security_start_timestamp);
2486                 dual_timestamp_serialize(f, "security-finish-timestamp", &m->security_finish_timestamp);
2487                 dual_timestamp_serialize(f, "generators-start-timestamp", &m->generators_start_timestamp);
2488                 dual_timestamp_serialize(f, "generators-finish-timestamp", &m->generators_finish_timestamp);
2489                 dual_timestamp_serialize(f, "units-load-start-timestamp", &m->units_load_start_timestamp);
2490                 dual_timestamp_serialize(f, "units-load-finish-timestamp", &m->units_load_finish_timestamp);
2491         }
2492
2493         if (!switching_root) {
2494                 STRV_FOREACH(e, m->environment) {
2495                         _cleanup_free_ char *ce;
2496
2497                         ce = cescape(*e);
2498                         if (!ce)
2499                                 return -ENOMEM;
2500
2501                         fprintf(f, "env=%s\n", *e);
2502                 }
2503         }
2504
2505         if (m->notify_fd >= 0) {
2506                 int copy;
2507
2508                 copy = fdset_put_dup(fds, m->notify_fd);
2509                 if (copy < 0)
2510                         return copy;
2511
2512                 fprintf(f, "notify-fd=%i\n", copy);
2513                 fprintf(f, "notify-socket=%s\n", m->notify_socket);
2514         }
2515
2516         if (m->cgroups_agent_fd >= 0) {
2517                 int copy;
2518
2519                 copy = fdset_put_dup(fds, m->cgroups_agent_fd);
2520                 if (copy < 0)
2521                         return copy;
2522
2523                 fprintf(f, "cgroups-agent-fd=%i\n", copy);
2524         }
2525
2526         if (m->user_lookup_fds[0] >= 0) {
2527                 int copy0, copy1;
2528
2529                 copy0 = fdset_put_dup(fds, m->user_lookup_fds[0]);
2530                 if (copy0 < 0)
2531                         return copy0;
2532
2533                 copy1 = fdset_put_dup(fds, m->user_lookup_fds[1]);
2534                 if (copy1 < 0)
2535                         return copy1;
2536
2537                 fprintf(f, "user-lookup=%i %i\n", copy0, copy1);
2538         }
2539
2540         bus_track_serialize(m->subscribed, f, "subscribed");
2541
2542         r = dynamic_user_serialize(m, f, fds);
2543         if (r < 0)
2544                 return r;
2545
2546         manager_serialize_uid_refs(m, f);
2547         manager_serialize_gid_refs(m, f);
2548
2549         fputc('\n', f);
2550
2551         HASHMAP_FOREACH_KEY(u, t, m->units, i) {
2552                 if (u->id != t)
2553                         continue;
2554
2555                 /* Start marker */
2556                 fputs(u->id, f);
2557                 fputc('\n', f);
2558
2559                 r = unit_serialize(u, f, fds, !switching_root);
2560                 if (r < 0) {
2561                         m->n_reloading--;
2562                         return r;
2563                 }
2564         }
2565
2566         assert(m->n_reloading > 0);
2567         m->n_reloading--;
2568
2569         if (ferror(f))
2570                 return -EIO;
2571
2572         r = bus_fdset_add_all(m, fds);
2573         if (r < 0)
2574                 return r;
2575
2576         return 0;
2577 }
2578
2579 int manager_deserialize(Manager *m, FILE *f, FDSet *fds) {
2580         int r = 0;
2581
2582         assert(m);
2583         assert(f);
2584
2585         log_debug("Deserializing state...");
2586
2587         m->n_reloading++;
2588
2589         for (;;) {
2590                 char line[LINE_MAX], *l;
2591                 const char *val;
2592
2593                 if (!fgets(line, sizeof(line), f)) {
2594                         if (feof(f))
2595                                 r = 0;
2596                         else
2597                                 r = -errno;
2598
2599                         goto finish;
2600                 }
2601
2602                 char_array_0(line);
2603                 l = strstrip(line);
2604
2605                 if (l[0] == 0)
2606                         break;
2607
2608                 if ((val = startswith(l, "current-job-id="))) {
2609                         uint32_t id;
2610
2611                         if (safe_atou32(val, &id) < 0)
2612                                 log_debug("Failed to parse current job id value %s", val);
2613                         else
2614                                 m->current_job_id = MAX(m->current_job_id, id);
2615
2616                 } else if ((val = startswith(l, "n-installed-jobs="))) {
2617                         uint32_t n;
2618
2619                         if (safe_atou32(val, &n) < 0)
2620                                 log_debug("Failed to parse installed jobs counter %s", val);
2621                         else
2622                                 m->n_installed_jobs += n;
2623
2624                 } else if ((val = startswith(l, "n-failed-jobs="))) {
2625                         uint32_t n;
2626
2627                         if (safe_atou32(val, &n) < 0)
2628                                 log_debug("Failed to parse failed jobs counter %s", val);
2629                         else
2630                                 m->n_failed_jobs += n;
2631
2632                 } else if ((val = startswith(l, "taint-usr="))) {
2633                         int b;
2634
2635                         b = parse_boolean(val);
2636                         if (b < 0)
2637                                 log_debug("Failed to parse taint /usr flag %s", val);
2638                         else
2639                                 m->taint_usr = m->taint_usr || b;
2640
2641                 } else if ((val = startswith(l, "firmware-timestamp=")))
2642                         dual_timestamp_deserialize(val, &m->firmware_timestamp);
2643                 else if ((val = startswith(l, "loader-timestamp=")))
2644                         dual_timestamp_deserialize(val, &m->loader_timestamp);
2645                 else if ((val = startswith(l, "kernel-timestamp=")))
2646                         dual_timestamp_deserialize(val, &m->kernel_timestamp);
2647                 else if ((val = startswith(l, "initrd-timestamp=")))
2648                         dual_timestamp_deserialize(val, &m->initrd_timestamp);
2649                 else if ((val = startswith(l, "userspace-timestamp=")))
2650                         dual_timestamp_deserialize(val, &m->userspace_timestamp);
2651                 else if ((val = startswith(l, "finish-timestamp=")))
2652                         dual_timestamp_deserialize(val, &m->finish_timestamp);
2653                 else if ((val = startswith(l, "security-start-timestamp=")))
2654                         dual_timestamp_deserialize(val, &m->security_start_timestamp);
2655                 else if ((val = startswith(l, "security-finish-timestamp=")))
2656                         dual_timestamp_deserialize(val, &m->security_finish_timestamp);
2657                 else if ((val = startswith(l, "generators-start-timestamp=")))
2658                         dual_timestamp_deserialize(val, &m->generators_start_timestamp);
2659                 else if ((val = startswith(l, "generators-finish-timestamp=")))
2660                         dual_timestamp_deserialize(val, &m->generators_finish_timestamp);
2661                 else if ((val = startswith(l, "units-load-start-timestamp=")))
2662                         dual_timestamp_deserialize(val, &m->units_load_start_timestamp);
2663                 else if ((val = startswith(l, "units-load-finish-timestamp=")))
2664                         dual_timestamp_deserialize(val, &m->units_load_finish_timestamp);
2665                 else if (startswith(l, "env=")) {
2666                         _cleanup_free_ char *uce = NULL;
2667                         char **e;
2668
2669                         r = cunescape(l + 4, UNESCAPE_RELAX, &uce);
2670                         if (r < 0)
2671                                 goto finish;
2672
2673                         e = strv_env_set(m->environment, uce);
2674                         if (!e) {
2675                                 r = -ENOMEM;
2676                                 goto finish;
2677                         }
2678
2679                         strv_free(m->environment);
2680                         m->environment = e;
2681
2682                 } else if ((val = startswith(l, "notify-fd="))) {
2683                         int fd;
2684
2685                         if (safe_atoi(val, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2686                                 log_debug("Failed to parse notify fd: %s", val);
2687                         else {
2688                                 m->notify_event_source = sd_event_source_unref(m->notify_event_source);
2689                                 safe_close(m->notify_fd);
2690                                 m->notify_fd = fdset_remove(fds, fd);
2691                         }
2692
2693                 } else if ((val = startswith(l, "notify-socket="))) {
2694                         char *n;
2695
2696                         n = strdup(val);
2697                         if (!n) {
2698                                 r = -ENOMEM;
2699                                 goto finish;
2700                         }
2701
2702                         free(m->notify_socket);
2703                         m->notify_socket = n;
2704
2705                 } else if ((val = startswith(l, "cgroups-agent-fd="))) {
2706                         int fd;
2707
2708                         if (safe_atoi(val, &fd) < 0 || fd < 0 || !fdset_contains(fds, fd))
2709                                 log_debug("Failed to parse cgroups agent fd: %s", val);
2710                         else {
2711                                 m->cgroups_agent_event_source = sd_event_source_unref(m->cgroups_agent_event_source);
2712                                 safe_close(m->cgroups_agent_fd);
2713                                 m->cgroups_agent_fd = fdset_remove(fds, fd);
2714                         }
2715
2716                 } else if ((val = startswith(l, "user-lookup="))) {
2717                         int fd0, fd1;
2718
2719                         if (sscanf(val, "%i %i", &fd0, &fd1) != 2 || fd0 < 0 || fd1 < 0 || fd0 == fd1 || !fdset_contains(fds, fd0) || !fdset_contains(fds, fd1))
2720                                 log_debug("Failed to parse user lookup fd: %s", val);
2721                         else {
2722                                 m->user_lookup_event_source = sd_event_source_unref(m->user_lookup_event_source);
2723                                 safe_close_pair(m->user_lookup_fds);
2724                                 m->user_lookup_fds[0] = fdset_remove(fds, fd0);
2725                                 m->user_lookup_fds[1] = fdset_remove(fds, fd1);
2726                         }
2727
2728                 } else if ((val = startswith(l, "dynamic-user=")))
2729                         dynamic_user_deserialize_one(m, val, fds);
2730                 else if ((val = startswith(l, "destroy-ipc-uid=")))
2731                         manager_deserialize_uid_refs_one(m, val);
2732                 else if ((val = startswith(l, "destroy-ipc-gid=")))
2733                         manager_deserialize_gid_refs_one(m, val);
2734                 else if ((val = startswith(l, "subscribed="))) {
2735
2736                         if (strv_extend(&m->deserialized_subscribed, val) < 0)
2737                                 log_oom();
2738
2739                 } else if (!startswith(l, "kdbus-fd=")) /* ignore this one */
2740                         log_debug("Unknown serialization item '%s'", l);
2741         }
2742
2743         for (;;) {
2744                 Unit *u;
2745                 char name[UNIT_NAME_MAX+2];
2746
2747                 /* Start marker */
2748                 if (!fgets(name, sizeof(name), f)) {
2749                         if (feof(f))
2750                                 r = 0;
2751                         else
2752                                 r = -errno;
2753
2754                         goto finish;
2755                 }
2756
2757                 char_array_0(name);
2758
2759                 r = manager_load_unit(m, strstrip(name), NULL, NULL, &u);
2760                 if (r < 0)
2761                         goto finish;
2762
2763                 r = unit_deserialize(u, f, fds);
2764                 if (r < 0)
2765                         goto finish;
2766         }
2767
2768 finish:
2769         if (ferror(f))
2770                 r = -EIO;
2771
2772         assert(m->n_reloading > 0);
2773         m->n_reloading--;
2774
2775         return r;
2776 }
2777
2778 int manager_reload(Manager *m) {
2779         int r, q;
2780         _cleanup_fclose_ FILE *f = NULL;
2781         _cleanup_fdset_free_ FDSet *fds = NULL;
2782
2783         assert(m);
2784
2785         r = manager_open_serialization(m, &f);
2786         if (r < 0)
2787                 return r;
2788
2789         m->n_reloading++;
2790         bus_manager_send_reloading(m, true);
2791
2792         fds = fdset_new();
2793         if (!fds) {
2794                 m->n_reloading--;
2795                 return -ENOMEM;
2796         }
2797
2798         r = manager_serialize(m, f, fds, false);
2799         if (r < 0) {
2800                 m->n_reloading--;
2801                 return r;
2802         }
2803
2804         if (fseeko(f, 0, SEEK_SET) < 0) {
2805                 m->n_reloading--;
2806                 return -errno;
2807         }
2808
2809         /* From here on there is no way back. */
2810         manager_clear_jobs_and_units(m);
2811         lookup_paths_flush_generator(&m->lookup_paths);
2812         lookup_paths_free(&m->lookup_paths);
2813         dynamic_user_vacuum(m, false);
2814         m->uid_refs = hashmap_free(m->uid_refs);
2815         m->gid_refs = hashmap_free(m->gid_refs);
2816
2817         q = lookup_paths_init(&m->lookup_paths, m->unit_file_scope, 0, NULL);
2818         if (q < 0 && r >= 0)
2819                 r = q;
2820
2821         /* Find new unit paths */
2822         q = manager_run_generators(m);
2823         if (q < 0 && r >= 0)
2824                 r = q;
2825
2826         lookup_paths_reduce(&m->lookup_paths);
2827         manager_build_unit_path_cache(m);
2828
2829         /* First, enumerate what we can from all config files */
2830         manager_enumerate(m);
2831
2832         /* Second, deserialize our stored data */
2833         q = manager_deserialize(m, f, fds);
2834         if (q < 0 && r >= 0)
2835                 r = q;
2836
2837         fclose(f);
2838         f = NULL;
2839
2840         /* Re-register notify_fd as event source */
2841         q = manager_setup_notify(m);
2842         if (q < 0 && r >= 0)
2843                 r = q;
2844
2845         q = manager_setup_cgroups_agent(m);
2846         if (q < 0 && r >= 0)
2847                 r = q;
2848
2849         q = manager_setup_user_lookup_fd(m);
2850         if (q < 0 && r >= 0)
2851                 r = q;
2852
2853         /* Third, fire things up! */
2854         manager_coldplug(m);
2855
2856         /* Release any dynamic users no longer referenced */
2857         dynamic_user_vacuum(m, true);
2858
2859         /* Release any references to UIDs/GIDs no longer referenced, and destroy any IPC owned by them */
2860         manager_vacuum_uid_refs(m);
2861         manager_vacuum_gid_refs(m);
2862
2863         /* Sync current state of bus names with our set of listening units */
2864         if (m->api_bus)
2865                 manager_sync_bus_names(m, m->api_bus);
2866
2867         assert(m->n_reloading > 0);
2868         m->n_reloading--;
2869
2870         m->send_reloading_done = true;
2871
2872         return r;
2873 }
2874
2875 void manager_reset_failed(Manager *m) {
2876         Unit *u;
2877         Iterator i;
2878
2879         assert(m);
2880
2881         HASHMAP_FOREACH(u, m->units, i)
2882                 unit_reset_failed(u);
2883 }
2884
2885 bool manager_unit_inactive_or_pending(Manager *m, const char *name) {
2886         Unit *u;
2887
2888         assert(m);
2889         assert(name);
2890
2891         /* Returns true if the unit is inactive or going down */
2892         u = manager_get_unit(m, name);
2893         if (!u)
2894                 return true;
2895
2896         return unit_inactive_or_pending(u);
2897 }
2898
2899 static void manager_notify_finished(Manager *m) {
2900         char userspace[FORMAT_TIMESPAN_MAX], initrd[FORMAT_TIMESPAN_MAX], kernel[FORMAT_TIMESPAN_MAX], sum[FORMAT_TIMESPAN_MAX];
2901         usec_t firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec;
2902
2903         if (m->test_run)
2904                 return;
2905
2906         if (MANAGER_IS_SYSTEM(m) && detect_container() <= 0) {
2907
2908                 /* Note that m->kernel_usec.monotonic is always at 0,
2909                  * and m->firmware_usec.monotonic and
2910                  * m->loader_usec.monotonic should be considered
2911                  * negative values. */
2912
2913                 firmware_usec = m->firmware_timestamp.monotonic - m->loader_timestamp.monotonic;
2914                 loader_usec = m->loader_timestamp.monotonic - m->kernel_timestamp.monotonic;
2915                 userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
2916                 total_usec = m->firmware_timestamp.monotonic + m->finish_timestamp.monotonic;
2917
2918                 if (dual_timestamp_is_set(&m->initrd_timestamp)) {
2919
2920                         kernel_usec = m->initrd_timestamp.monotonic - m->kernel_timestamp.monotonic;
2921                         initrd_usec = m->userspace_timestamp.monotonic - m->initrd_timestamp.monotonic;
2922
2923                         log_struct(LOG_INFO,
2924                                    LOG_MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED),
2925                                    "KERNEL_USEC="USEC_FMT, kernel_usec,
2926                                    "INITRD_USEC="USEC_FMT, initrd_usec,
2927                                    "USERSPACE_USEC="USEC_FMT, userspace_usec,
2928                                    LOG_MESSAGE("Startup finished in %s (kernel) + %s (initrd) + %s (userspace) = %s.",
2929                                                format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
2930                                                format_timespan(initrd, sizeof(initrd), initrd_usec, USEC_PER_MSEC),
2931                                                format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
2932                                                format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
2933                                    NULL);
2934                 } else {
2935                         kernel_usec = m->userspace_timestamp.monotonic - m->kernel_timestamp.monotonic;
2936                         initrd_usec = 0;
2937
2938                         log_struct(LOG_INFO,
2939                                    LOG_MESSAGE_ID(SD_MESSAGE_STARTUP_FINISHED),
2940                                    "KERNEL_USEC="USEC_FMT, kernel_usec,
2941                                    "USERSPACE_USEC="USEC_FMT, userspace_usec,
2942                                    LOG_MESSAGE("Startup finished in %s (kernel) + %s (userspace) = %s.",
2943                                                format_timespan(kernel, sizeof(kernel), kernel_usec, USEC_PER_MSEC),
2944                                                format_timespan(userspace, sizeof(userspace), userspace_usec, USEC_PER_MSEC),
2945                                                format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
2946                                    NULL);
2947                 }
2948         } else {
2949                 firmware_usec = loader_usec = initrd_usec = kernel_usec = 0;
2950                 total_usec = userspace_usec = m->finish_timestamp.monotonic - m->userspace_timestamp.monotonic;
2951
2952                 log_struct(LOG_INFO,
2953                            LOG_MESSAGE_ID(SD_MESSAGE_USER_STARTUP_FINISHED),
2954                            "USERSPACE_USEC="USEC_FMT, userspace_usec,
2955                            LOG_MESSAGE("Startup finished in %s.",
2956                                        format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC)),
2957                            NULL);
2958         }
2959
2960         bus_manager_send_finished(m, firmware_usec, loader_usec, kernel_usec, initrd_usec, userspace_usec, total_usec);
2961
2962         sd_notifyf(false,
2963                    "READY=1\n"
2964                    "STATUS=Startup finished in %s.",
2965                    format_timespan(sum, sizeof(sum), total_usec, USEC_PER_MSEC));
2966 }
2967
2968 void manager_check_finished(Manager *m) {
2969         assert(m);
2970
2971         if (MANAGER_IS_RELOADING(m))
2972                 return;
2973
2974         /* Verify that we are actually running currently. Initially
2975          * the exit code is set to invalid, and during operation it is
2976          * then set to MANAGER_OK */
2977         if (m->exit_code != MANAGER_OK)
2978                 return;
2979
2980         if (hashmap_size(m->jobs) > 0) {
2981                 if (m->jobs_in_progress_event_source)
2982                         /* Ignore any failure, this is only for feedback */
2983                         (void) sd_event_source_set_time(m->jobs_in_progress_event_source, now(CLOCK_MONOTONIC) + JOBS_IN_PROGRESS_WAIT_USEC);
2984
2985                 return;
2986         }
2987
2988         manager_flip_auto_status(m, false);
2989
2990         /* Notify Type=idle units that we are done now */
2991         manager_close_idle_pipe(m);
2992
2993         /* Turn off confirm spawn now */
2994         m->confirm_spawn = NULL;
2995
2996         /* No need to update ask password status when we're going non-interactive */
2997         manager_close_ask_password(m);
2998
2999         /* This is no longer the first boot */
3000         manager_set_first_boot(m, false);
3001
3002         if (dual_timestamp_is_set(&m->finish_timestamp))
3003                 return;
3004
3005         dual_timestamp_get(&m->finish_timestamp);
3006
3007         manager_notify_finished(m);
3008
3009         manager_invalidate_startup_units(m);
3010 }
3011
3012 static int manager_run_generators(Manager *m) {
3013         _cleanup_strv_free_ char **paths = NULL;
3014         const char *argv[5];
3015         char **path;
3016         int r;
3017
3018         assert(m);
3019
3020         if (m->test_run)
3021                 return 0;
3022
3023         paths = generator_binary_paths(m->unit_file_scope);
3024         if (!paths)
3025                 return log_oom();
3026
3027         /* Optimize by skipping the whole process by not creating output directories
3028          * if no generators are found. */
3029         STRV_FOREACH(path, paths) {
3030                 if (access(*path, F_OK) >= 0)
3031                         goto found;
3032                 if (errno != ENOENT)
3033                         log_warning_errno(errno, "Failed to open generator directory %s: %m", *path);
3034         }
3035
3036         return 0;
3037
3038  found:
3039         r = lookup_paths_mkdir_generator(&m->lookup_paths);
3040         if (r < 0)
3041                 goto finish;
3042
3043         argv[0] = NULL; /* Leave this empty, execute_directory() will fill something in */
3044         argv[1] = m->lookup_paths.generator;
3045         argv[2] = m->lookup_paths.generator_early;
3046         argv[3] = m->lookup_paths.generator_late;
3047         argv[4] = NULL;
3048
3049         RUN_WITH_UMASK(0022)
3050                 execute_directories((const char* const*) paths, DEFAULT_TIMEOUT_USEC, (char**) argv);
3051
3052 finish:
3053         lookup_paths_trim_generator(&m->lookup_paths);
3054         return r;
3055 }
3056
3057 int manager_environment_add(Manager *m, char **minus, char **plus) {
3058         char **a = NULL, **b = NULL, **l;
3059         assert(m);
3060
3061         l = m->environment;
3062
3063         if (!strv_isempty(minus)) {
3064                 a = strv_env_delete(l, 1, minus);
3065                 if (!a)
3066                         return -ENOMEM;
3067
3068                 l = a;
3069         }
3070
3071         if (!strv_isempty(plus)) {
3072                 b = strv_env_merge(2, l, plus);
3073                 if (!b) {
3074                         strv_free(a);
3075                         return -ENOMEM;
3076                 }
3077
3078                 l = b;
3079         }
3080
3081         if (m->environment != l)
3082                 strv_free(m->environment);
3083         if (a != l)
3084                 strv_free(a);
3085         if (b != l)
3086                 strv_free(b);
3087
3088         m->environment = l;
3089         manager_clean_environment(m);
3090         strv_sort(m->environment);
3091
3092         return 0;
3093 }
3094
3095 int manager_set_default_rlimits(Manager *m, struct rlimit **default_rlimit) {
3096         int i;
3097
3098         assert(m);
3099
3100         for (i = 0; i < _RLIMIT_MAX; i++) {
3101                 m->rlimit[i] = mfree(m->rlimit[i]);
3102
3103                 if (!default_rlimit[i])
3104                         continue;
3105
3106                 m->rlimit[i] = newdup(struct rlimit, default_rlimit[i], 1);
3107                 if (!m->rlimit[i])
3108                         return log_oom();
3109         }
3110
3111         return 0;
3112 }
3113
3114 void manager_recheck_journal(Manager *m) {
3115         Unit *u;
3116
3117         assert(m);
3118
3119         if (!MANAGER_IS_SYSTEM(m))
3120                 return;
3121
3122         u = manager_get_unit(m, SPECIAL_JOURNALD_SOCKET);
3123         if (u && SOCKET(u)->state != SOCKET_RUNNING) {
3124                 log_close_journal();
3125                 return;
3126         }
3127
3128         u = manager_get_unit(m, SPECIAL_JOURNALD_SERVICE);
3129         if (u && SERVICE(u)->state != SERVICE_RUNNING) {
3130                 log_close_journal();
3131                 return;
3132         }
3133
3134         /* Hmm, OK, so the socket is fully up and the service is up
3135          * too, then let's make use of the thing. */
3136         log_open();
3137 }
3138
3139 void manager_set_show_status(Manager *m, ShowStatus mode) {
3140         assert(m);
3141         assert(IN_SET(mode, SHOW_STATUS_AUTO, SHOW_STATUS_NO, SHOW_STATUS_YES, SHOW_STATUS_TEMPORARY));
3142
3143         if (!MANAGER_IS_SYSTEM(m))
3144                 return;
3145
3146         if (m->show_status != mode)
3147                 log_debug("%s showing of status.",
3148                           mode == SHOW_STATUS_NO ? "Disabling" : "Enabling");
3149         m->show_status = mode;
3150
3151         if (mode > 0)
3152                 (void) touch("/run/systemd/show-status");
3153         else
3154                 (void) unlink("/run/systemd/show-status");
3155 }
3156
3157 static bool manager_get_show_status(Manager *m, StatusType type) {
3158         assert(m);
3159
3160         if (!MANAGER_IS_SYSTEM(m))
3161                 return false;
3162
3163         if (m->no_console_output)
3164                 return false;
3165
3166         if (!IN_SET(manager_state(m), MANAGER_INITIALIZING, MANAGER_STARTING, MANAGER_STOPPING))
3167                 return false;
3168
3169         /* If we cannot find out the status properly, just proceed. */
3170         if (type != STATUS_TYPE_EMERGENCY && manager_check_ask_password(m) > 0)
3171                 return false;
3172
3173         if (m->show_status > 0)
3174                 return true;
3175
3176         return false;
3177 }
3178
3179 const char *manager_get_confirm_spawn(Manager *m) {
3180         static int last_errno = 0;
3181         const char *vc = m->confirm_spawn;
3182         struct stat st;
3183         int r;
3184
3185         /* Here's the deal: we want to test the validity of the console but don't want
3186          * PID1 to go through the whole console process which might block. But we also
3187          * want to warn the user only once if something is wrong with the console so we
3188          * cannot do the sanity checks after spawning our children. So here we simply do
3189          * really basic tests to hopefully trap common errors.
3190          *
3191          * If the console suddenly disappear at the time our children will really it
3192          * then they will simply fail to acquire it and a positive answer will be
3193          * assumed. New children will fallback to /dev/console though.
3194          *
3195          * Note: TTYs are devices that can come and go any time, and frequently aren't
3196          * available yet during early boot (consider a USB rs232 dongle...). If for any
3197          * reason the configured console is not ready, we fallback to the default
3198          * console. */
3199
3200         if (!vc || path_equal(vc, "/dev/console"))
3201                 return vc;
3202
3203         r = stat(vc, &st);
3204         if (r < 0)
3205                 goto fail;
3206
3207         if (!S_ISCHR(st.st_mode)) {
3208                 errno = ENOTTY;
3209                 goto fail;
3210         }
3211
3212         last_errno = 0;
3213         return vc;
3214 fail:
3215         if (last_errno != errno) {
3216                 last_errno = errno;
3217                 log_warning_errno(errno, "Failed to open %s: %m, using default console", vc);
3218         }
3219         return "/dev/console";
3220 }
3221
3222 void manager_set_first_boot(Manager *m, bool b) {
3223         assert(m);
3224
3225         if (!MANAGER_IS_SYSTEM(m))
3226                 return;
3227
3228         if (m->first_boot != (int) b) {
3229                 if (b)
3230                         (void) touch("/run/systemd/first-boot");
3231                 else
3232                         (void) unlink("/run/systemd/first-boot");
3233         }
3234
3235         m->first_boot = b;
3236 }
3237
3238 void manager_disable_confirm_spawn(void) {
3239         (void) touch("/run/systemd/confirm_spawn_disabled");
3240 }
3241
3242 bool manager_is_confirm_spawn_disabled(Manager *m) {
3243         if (!m->confirm_spawn)
3244                 return true;
3245
3246         return access("/run/systemd/confirm_spawn_disabled", F_OK) >= 0;
3247 }
3248
3249 void manager_status_printf(Manager *m, StatusType type, const char *status, const char *format, ...) {
3250         va_list ap;
3251
3252         /* If m is NULL, assume we're after shutdown and let the messages through. */
3253
3254         if (m && !manager_get_show_status(m, type))
3255                 return;
3256
3257         /* XXX We should totally drop the check for ephemeral here
3258          * and thus effectively make 'Type=idle' pointless. */
3259         if (type == STATUS_TYPE_EPHEMERAL && m && m->n_on_console > 0)
3260                 return;
3261
3262         va_start(ap, format);
3263         status_vprintf(status, true, type == STATUS_TYPE_EPHEMERAL, format, ap);
3264         va_end(ap);
3265 }
3266
3267 Set *manager_get_units_requiring_mounts_for(Manager *m, const char *path) {
3268         char p[strlen(path)+1];
3269
3270         assert(m);
3271         assert(path);
3272
3273         strcpy(p, path);
3274         path_kill_slashes(p);
3275
3276         return hashmap_get(m->units_requiring_mounts_for, streq(p, "/") ? "" : p);
3277 }
3278
3279 const char *manager_get_runtime_prefix(Manager *m) {
3280         assert(m);
3281
3282         return MANAGER_IS_SYSTEM(m) ?
3283                "/run" :
3284                getenv("XDG_RUNTIME_DIR");
3285 }
3286
3287 int manager_update_failed_units(Manager *m, Unit *u, bool failed) {
3288         unsigned size;
3289         int r;
3290
3291         assert(m);
3292         assert(u->manager == m);
3293
3294         size = set_size(m->failed_units);
3295
3296         if (failed) {
3297                 r = set_ensure_allocated(&m->failed_units, NULL);
3298                 if (r < 0)
3299                         return log_oom();
3300
3301                 if (set_put(m->failed_units, u) < 0)
3302                         return log_oom();
3303         } else
3304                 (void) set_remove(m->failed_units, u);
3305
3306         if (set_size(m->failed_units) != size)
3307                 bus_manager_send_change_signal(m);
3308
3309         return 0;
3310 }
3311
3312 ManagerState manager_state(Manager *m) {
3313         Unit *u;
3314
3315         assert(m);
3316
3317         /* Did we ever finish booting? If not then we are still starting up */
3318         if (!dual_timestamp_is_set(&m->finish_timestamp)) {
3319
3320                 u = manager_get_unit(m, SPECIAL_BASIC_TARGET);
3321                 if (!u || !UNIT_IS_ACTIVE_OR_RELOADING(unit_active_state(u)))
3322                         return MANAGER_INITIALIZING;
3323
3324                 return MANAGER_STARTING;
3325         }
3326
3327         /* Is the special shutdown target queued? If so, we are in shutdown state */
3328         u = manager_get_unit(m, SPECIAL_SHUTDOWN_TARGET);
3329         if (u && u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))
3330                 return MANAGER_STOPPING;
3331
3332         /* Are the rescue or emergency targets active or queued? If so we are in maintenance state */
3333         u = manager_get_unit(m, SPECIAL_RESCUE_TARGET);
3334         if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
3335                   (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))))
3336                 return MANAGER_MAINTENANCE;
3337
3338         u = manager_get_unit(m, SPECIAL_EMERGENCY_TARGET);
3339         if (u && (UNIT_IS_ACTIVE_OR_ACTIVATING(unit_active_state(u)) ||
3340                   (u->job && IN_SET(u->job->type, JOB_START, JOB_RESTART, JOB_RELOAD_OR_START))))
3341                 return MANAGER_MAINTENANCE;
3342
3343         /* Are there any failed units? If so, we are in degraded mode */
3344         if (set_size(m->failed_units) > 0)
3345                 return MANAGER_DEGRADED;
3346
3347         return MANAGER_RUNNING;
3348 }
3349
3350 #define DESTROY_IPC_FLAG (UINT32_C(1) << 31)
3351
3352 static void manager_unref_uid_internal(
3353                 Manager *m,
3354                 Hashmap **uid_refs,
3355                 uid_t uid,
3356                 bool destroy_now,
3357                 int (*_clean_ipc)(uid_t uid)) {
3358
3359         uint32_t c, n;
3360
3361         assert(m);
3362         assert(uid_refs);
3363         assert(uid_is_valid(uid));
3364         assert(_clean_ipc);
3365
3366         /* A generic implementation, covering both manager_unref_uid() and manager_unref_gid(), under the assumption
3367          * that uid_t and gid_t are actually defined the same way, with the same validity rules.
3368          *
3369          * We store a hashmap where the UID/GID is they key and the value is a 32bit reference counter, whose highest
3370          * bit is used as flag for marking UIDs/GIDs whose IPC objects to remove when the last reference to the UID/GID
3371          * is dropped. The flag is set to on, once at least one reference from a unit where RemoveIPC= is set is added
3372          * on a UID/GID. It is reset when the UID's/GID's reference counter drops to 0 again. */
3373
3374         assert_cc(sizeof(uid_t) == sizeof(gid_t));
3375         assert_cc(UID_INVALID == (uid_t) GID_INVALID);
3376
3377         if (uid == 0) /* We don't keep track of root, and will never destroy it */
3378                 return;
3379
3380         c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3381
3382         n = c & ~DESTROY_IPC_FLAG;
3383         assert(n > 0);
3384         n--;
3385
3386         if (destroy_now && n == 0) {
3387                 hashmap_remove(*uid_refs, UID_TO_PTR(uid));
3388
3389                 if (c & DESTROY_IPC_FLAG) {
3390                         log_debug("%s " UID_FMT " is no longer referenced, cleaning up its IPC.",
3391                                   _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
3392                                   uid);
3393                         (void) _clean_ipc(uid);
3394                 }
3395         } else {
3396                 c = n | (c & DESTROY_IPC_FLAG);
3397                 assert_se(hashmap_update(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c)) >= 0);
3398         }
3399 }
3400
3401 void manager_unref_uid(Manager *m, uid_t uid, bool destroy_now) {
3402         manager_unref_uid_internal(m, &m->uid_refs, uid, destroy_now, clean_ipc_by_uid);
3403 }
3404
3405 void manager_unref_gid(Manager *m, gid_t gid, bool destroy_now) {
3406         manager_unref_uid_internal(m, &m->gid_refs, (uid_t) gid, destroy_now, clean_ipc_by_gid);
3407 }
3408
3409 static int manager_ref_uid_internal(
3410                 Manager *m,
3411                 Hashmap **uid_refs,
3412                 uid_t uid,
3413                 bool clean_ipc) {
3414
3415         uint32_t c, n;
3416         int r;
3417
3418         assert(m);
3419         assert(uid_refs);
3420         assert(uid_is_valid(uid));
3421
3422         /* A generic implementation, covering both manager_ref_uid() and manager_ref_gid(), under the assumption
3423          * that uid_t and gid_t are actually defined the same way, with the same validity rules. */
3424
3425         assert_cc(sizeof(uid_t) == sizeof(gid_t));
3426         assert_cc(UID_INVALID == (uid_t) GID_INVALID);
3427
3428         if (uid == 0) /* We don't keep track of root, and will never destroy it */
3429                 return 0;
3430
3431         r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
3432         if (r < 0)
3433                 return r;
3434
3435         c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3436
3437         n = c & ~DESTROY_IPC_FLAG;
3438         n++;
3439
3440         if (n & DESTROY_IPC_FLAG) /* check for overflow */
3441                 return -EOVERFLOW;
3442
3443         c = n | (c & DESTROY_IPC_FLAG) | (clean_ipc ? DESTROY_IPC_FLAG : 0);
3444
3445         return hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
3446 }
3447
3448 int manager_ref_uid(Manager *m, uid_t uid, bool clean_ipc) {
3449         return manager_ref_uid_internal(m, &m->uid_refs, uid, clean_ipc);
3450 }
3451
3452 int manager_ref_gid(Manager *m, gid_t gid, bool clean_ipc) {
3453         return manager_ref_uid_internal(m, &m->gid_refs, (uid_t) gid, clean_ipc);
3454 }
3455
3456 static void manager_vacuum_uid_refs_internal(
3457                 Manager *m,
3458                 Hashmap **uid_refs,
3459                 int (*_clean_ipc)(uid_t uid)) {
3460
3461         Iterator i;
3462         void *p, *k;
3463
3464         assert(m);
3465         assert(uid_refs);
3466         assert(_clean_ipc);
3467
3468         HASHMAP_FOREACH_KEY(p, k, *uid_refs, i) {
3469                 uint32_t c, n;
3470                 uid_t uid;
3471
3472                 uid = PTR_TO_UID(k);
3473                 c = PTR_TO_UINT32(p);
3474
3475                 n = c & ~DESTROY_IPC_FLAG;
3476                 if (n > 0)
3477                         continue;
3478
3479                 if (c & DESTROY_IPC_FLAG) {
3480                         log_debug("Found unreferenced %s " UID_FMT " after reload/reexec. Cleaning up.",
3481                                   _clean_ipc == clean_ipc_by_uid ? "UID" : "GID",
3482                                   uid);
3483                         (void) _clean_ipc(uid);
3484                 }
3485
3486                 assert_se(hashmap_remove(*uid_refs, k) == p);
3487         }
3488 }
3489
3490 void manager_vacuum_uid_refs(Manager *m) {
3491         manager_vacuum_uid_refs_internal(m, &m->uid_refs, clean_ipc_by_uid);
3492 }
3493
3494 void manager_vacuum_gid_refs(Manager *m) {
3495         manager_vacuum_uid_refs_internal(m, &m->gid_refs, clean_ipc_by_gid);
3496 }
3497
3498 static void manager_serialize_uid_refs_internal(
3499                 Manager *m,
3500                 FILE *f,
3501                 Hashmap **uid_refs,
3502                 const char *field_name) {
3503
3504         Iterator i;
3505         void *p, *k;
3506
3507         assert(m);
3508         assert(f);
3509         assert(uid_refs);
3510         assert(field_name);
3511
3512         /* Serialize the UID reference table. Or actually, just the IPC destruction flag of it, as the actual counter
3513          * of it is better rebuild after a reload/reexec. */
3514
3515         HASHMAP_FOREACH_KEY(p, k, *uid_refs, i) {
3516                 uint32_t c;
3517                 uid_t uid;
3518
3519                 uid = PTR_TO_UID(k);
3520                 c = PTR_TO_UINT32(p);
3521
3522                 if (!(c & DESTROY_IPC_FLAG))
3523                         continue;
3524
3525                 fprintf(f, "%s=" UID_FMT "\n", field_name, uid);
3526         }
3527 }
3528
3529 void manager_serialize_uid_refs(Manager *m, FILE *f) {
3530         manager_serialize_uid_refs_internal(m, f, &m->uid_refs, "destroy-ipc-uid");
3531 }
3532
3533 void manager_serialize_gid_refs(Manager *m, FILE *f) {
3534         manager_serialize_uid_refs_internal(m, f, &m->gid_refs, "destroy-ipc-gid");
3535 }
3536
3537 static void manager_deserialize_uid_refs_one_internal(
3538                 Manager *m,
3539                 Hashmap** uid_refs,
3540                 const char *value) {
3541
3542         uid_t uid;
3543         uint32_t c;
3544         int r;
3545
3546         assert(m);
3547         assert(uid_refs);
3548         assert(value);
3549
3550         r = parse_uid(value, &uid);
3551         if (r < 0 || uid == 0) {
3552                 log_debug("Unable to parse UID reference serialization");
3553                 return;
3554         }
3555
3556         r = hashmap_ensure_allocated(uid_refs, &trivial_hash_ops);
3557         if (r < 0) {
3558                 log_oom();
3559                 return;
3560         }
3561
3562         c = PTR_TO_UINT32(hashmap_get(*uid_refs, UID_TO_PTR(uid)));
3563         if (c & DESTROY_IPC_FLAG)
3564                 return;
3565
3566         c |= DESTROY_IPC_FLAG;
3567
3568         r = hashmap_replace(*uid_refs, UID_TO_PTR(uid), UINT32_TO_PTR(c));
3569         if (r < 0) {
3570                 log_debug("Failed to add UID reference entry");
3571                 return;
3572         }
3573 }
3574
3575 void manager_deserialize_uid_refs_one(Manager *m, const char *value) {
3576         manager_deserialize_uid_refs_one_internal(m, &m->uid_refs, value);
3577 }
3578
3579 void manager_deserialize_gid_refs_one(Manager *m, const char *value) {
3580         manager_deserialize_uid_refs_one_internal(m, &m->gid_refs, value);
3581 }
3582
3583 int manager_dispatch_user_lookup_fd(sd_event_source *source, int fd, uint32_t revents, void *userdata) {
3584         struct buffer {
3585                 uid_t uid;
3586                 gid_t gid;
3587                 char unit_name[UNIT_NAME_MAX+1];
3588         } _packed_ buffer;
3589
3590         Manager *m = userdata;
3591         ssize_t l;
3592         size_t n;
3593         Unit *u;
3594
3595         assert_se(source);
3596         assert_se(m);
3597
3598         /* Invoked whenever a child process succeeded resolving its user/group to use and sent us the resulting UID/GID
3599          * in a datagram. We parse the datagram here and pass it off to the unit, so that it can add a reference to the
3600          * UID/GID so that it can destroy the UID/GID's IPC objects when the reference counter drops to 0. */
3601
3602         l = recv(fd, &buffer, sizeof(buffer), MSG_DONTWAIT);
3603         if (l < 0) {
3604                 if (errno == EINTR || errno == EAGAIN)
3605                         return 0;
3606
3607                 return log_error_errno(errno, "Failed to read from user lookup fd: %m");
3608         }
3609
3610         if ((size_t) l <= offsetof(struct buffer, unit_name)) {
3611                 log_warning("Received too short user lookup message, ignoring.");
3612                 return 0;
3613         }
3614
3615         if ((size_t) l > offsetof(struct buffer, unit_name) + UNIT_NAME_MAX) {
3616                 log_warning("Received too long user lookup message, ignoring.");
3617                 return 0;
3618         }
3619
3620         if (!uid_is_valid(buffer.uid) && !gid_is_valid(buffer.gid)) {
3621                 log_warning("Got user lookup message with invalid UID/GID pair, ignoring.");
3622                 return 0;
3623         }
3624
3625         n = (size_t) l - offsetof(struct buffer, unit_name);
3626         if (memchr(buffer.unit_name, 0, n)) {
3627                 log_warning("Received lookup message with embedded NUL character, ignoring.");
3628                 return 0;
3629         }
3630
3631         buffer.unit_name[n] = 0;
3632         u = manager_get_unit(m, buffer.unit_name);
3633         if (!u) {
3634                 log_debug("Got user lookup message but unit doesn't exist, ignoring.");
3635                 return 0;
3636         }
3637
3638         log_unit_debug(u, "User lookup succeeded: uid=" UID_FMT " gid=" GID_FMT, buffer.uid, buffer.gid);
3639
3640         unit_notify_user_lookup(u, buffer.uid, buffer.gid);
3641         return 0;
3642 }
3643
3644 static const char *const manager_state_table[_MANAGER_STATE_MAX] = {
3645         [MANAGER_INITIALIZING] = "initializing",
3646         [MANAGER_STARTING] = "starting",
3647         [MANAGER_RUNNING] = "running",
3648         [MANAGER_DEGRADED] = "degraded",
3649         [MANAGER_MAINTENANCE] = "maintenance",
3650         [MANAGER_STOPPING] = "stopping",
3651 };
3652
3653 DEFINE_STRING_TABLE_LOOKUP(manager_state, ManagerState);